7.0: managed ZIP writer, parallel record writes, checksum verification, and Excel-conformance fixes

[GabrielMarquezMatte]

What does this change?

Everything on develop since #140, completing the candidate for 7.0.0: 41 commits, 106 files.
The write path gets its own ZIP container and deflate encoder, records can be written on several
cores, the reader verifies ZIP checksums, and three bug hunts against real Excel fixed 17 cases
where the library produced or read something Excel disagrees with. The v6 → v7 port, including the
behavior changes below, is in docs/guide/migrating-to-v7.md.

The C ABI keeps every signature and XL_ABI_VERSION stays 6. xl_open_options gains one field in
what was tail padding, so its size is unchanged.

Managed ZIP writer

• XLSX and XLSB packages are written by Writer/Zip/ZipPackageWriter instead of ZipArchive. It
  writes the same layout (patched local headers on seekable output, data descriptors otherwise, Zip64
  only when a field overflows).
• CompressionLevel.Fastest, the default, uses a managed fixed-Huffman deflate encoder with a
  stored-block fallback and a CRC-32 that folds with PCLMULQDQ. Optimal and SmallestSize keep zlib
  inside the same container.
• 50,000 rows by four columns into a MemoryStream (Ryzen 7 5700X): XLSX 10.43 → 9.11 ms, XLSB
  7.66 → 5.42 ms; with PrefetchWrite, XLSB 5.37 → 3.08 ms. File sizes stay within 0.2%.
• A write that faults still leaves a central directory for the entries that completed. A workbook
  torn down mid-row with PrefetchWrite no longer risks a crash: the background writer is stopped
  before the package is abandoned.
• New deflate and deflate-segments fuzz targets check the encoder against two inflaters.

Parallel record writes (new API)

• WriteRecordsParallelAsync and WriteTableParallelAsync on XlsxSheetWriter and
  XlsbSheetWriter, with ParallelWriteOptions (DegreeOfParallelism, BatchRows). Records are cut
  into batches; each batch is formatted and compressed on a pool thread into a self-contained deflate
  segment, and segments are appended to the sheet in order.
• 1,000,000 rows by four columns at Fastest: XLSX 187 ms serial, 50 ms with 4 batches in flight,
  36 ms with 8; XLSB 124 / 41 / 28 ms. The file grows by 0.18% (XLSX) or 0.04% (XLSB).
• Memory is about 3 MB per batch in flight, whatever the sheet's size.
• The row delegate runs on several threads at once. A failure or a cancellation ends the sheet and
  its workbook. With UseSharedStrings, string indexes depend on thread timing, and many distinct
  strings (tens of thousands) limit the speed-up.

ZIP checksums on read (behavior change)

• Each ZIP entry of an XLSX or XLSB package is checked against its CRC-32. A mismatch throws
  InvalidDataException when the entry's last bytes are read. v6 checked only the entry's size, so
  a damaged sheet could read as plausible but wrong cells.
• ExcelReaderOptions.VerifyZipChecksums (default true) turns it off; the bindings expose it as
  verify_zip_checksums.
• It costs 2% to 4% of a read, and under 1% with PrefetchDecompression, where it runs on the
  background thread.

XLS writer

• Numbers that fit are written as RK and MULRK records, and long text goes through a shared-string
  table, so it is stored once.
• The worksheet substream follows the row-block layout Excel expects, with conformance tests on the
  record structure.

Fixes found by comparing against Excel 16

Readers

• XLSX text with a line break reads with \n, not \r\n. Excel stores the break as a CR LF pair in
  the XML; XLSB and XLS already returned \n.
• An XLSX formula whose text result begins or ends with whitespace no longer reads as empty
  (<v xml:space="preserve">).
• An XLS formula filled down or entered as an array no longer loses the text result of its first cell.

Writers (each of these produced a file Excel refused or altered)

• Sheet names with a leading or trailing apostrophe or a control character, column settings past the
  format's last column, a column width above 255, more than 200 distinct number formats and a number
  format longer than 255 characters now throw.
• In XLSX, a carriage return, U+FFFE and U+FFFF in cell text are written escaped.
• An XLSB sheet stops at 1,048,576 rows; it accepted one more.
• Write<T> with a type that has no numeric form throws ArgumentException in every writer; XLSB
  and XLS threw InvalidCastException.

Typed parser

• A numeric 0 or 1 binds to bool in XLSB and XLS, as it did in XLSX. A 0/1 flag column read from
  those formats came back all false.
• TimeOnly binds from a text cell in every format, and numeric cells in XLSB and XLS bind to
  TimeSpan, other parsable types and enums as they do in XLSX.

Bindings

• Timestamp text with Z or an offset no longer shifts by the machine's time zone when a CSV column
  is read as a timestamp.
• Python: an int outside the signed 64-bit range raises OverflowError instead of being written
  as its low 64 bits, and Cell.as_date is right for January and February 1900.

Other

• ExcelLimitExceededException.Message no longer says "bytes" for limits counted in rows, columns
  or entries.
• The static PGO profile for NativeAOT is regenerated.

How was it verified?

• Every writer fix was confirmed by opening the output in Excel 16 through COM; the reader fixes by
  comparing the library's cells with Excel's own values for workbooks Excel saved in XLSX, XLSB and
  XLS, in both date systems.
• Parallel reads were compared with sequential reads on seven 60,000-row files at six degrees of
  parallelism; parallel writes were opened in Excel at 1,000,000 rows and checked with Python's
  zipfile.testzip().
• The Python and Rust suites pass against a freshly published native library.

Known gaps

• The C++ binding was not built or run locally; CI covers it.
• The read tables in docs/performance/benchmarks.md predate checksum verification and are 2% to 4%
  optimistic for XLSX and XLSB.
• The PGO trainer covers reads only, so the new write path has no static profile under NativeAOT.
• Native error messages still carry resource keys (Arg_ParamName_Name) after the text.
• Parallel record writes are not exposed through the C ABI.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 93.21368% with 121 lines in your changes missing coverage. Please review.
✅ Project coverage is 91.04%. Comparing base (bcc628f) to head (6c46480).

Files with missing lines	Patch %	Lines
...eader.Core/Writer/ParallelSheetWriterExtensions.cs	50.00%	24 Missing ⚠️
src/ExcelReader.Core/Writer/Zip/ZipEntryStream.cs	91.54%	15 Missing and 3 partials ⚠️
src/ExcelReader.Core/Writer/Zip/Crc32.cs	91.73%	9 Missing and 1 partial ⚠️
...rc/ExcelReader.Core/Writer/Xlsb/XlsbSheetWriter.cs	91.17%	6 Missing and 3 partials ⚠️
src/ExcelReader.Core/Writer/Zip/BufferSink.cs	55.55%	8 Missing ⚠️
...lReader.Core/Writer/Internal/WriteOffloadStream.cs	64.70%	4 Missing and 2 partials ⚠️
...rc/ExcelReader.Core/Writer/Xls/XlsSharedStrings.cs	95.32%	1 Missing and 4 partials ⚠️
...rc/ExcelReader.Core/Writer/Xlsx/XlsxSheetWriter.cs	93.15%	4 Missing and 1 partial ⚠️
.../ExcelReader.Core/Writer/Internal/CellFormatter.cs	94.20%	2 Missing and 2 partials ⚠️
...eader.Core/Writer/Parallel/ParallelRecordWriter.cs	91.48%	3 Missing and 1 partial ⚠️
... and 10 more
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #141      +/-   ##
==========================================
+ Coverage   90.58%   91.04%   +0.45%     
==========================================
  Files         232      251      +19     
  Lines       15981    17472    +1491     
  Branches     2934     3153     +219     
==========================================
+ Hits        14477    15907    +1430     
- Misses       1016     1073      +57     
- Partials      488      492       +4     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.