Preserve quoted-printable escapes across write boundaries

[AryanRG]

A valid quoted-printable stream can silently decode differently depending on upload chunking. When a write ends with =3D=, the current decoder retains the last two bytes instead of caching from the incomplete equals sign. It splits a completed escape, so subsequent writes cannot recover the original bytes. For example, =3D=3D should decode to == regardless of where it is split.

Cache from the position of the incomplete escape in the final two bytes, leaving completed preceding escapes intact. Add decoder and public FormParser field/file regressions across every two-write split, including soft line breaks.

Validation:

• Six new regression cases failed on pristine main and pass with this change.
• Full local suite: 165 passed, with 100% statement coverage under repository exclusions.
• Ruff and mypy passed.
• 22,278 valid-stream/chunking comparisons matched one-shot binascii decoding after the fix; the base differed in 4,219.
• Targeted decoder and callback checks passed on Python 3.11 through 3.15. Full hosted CI and platform matrix remain unverified.

Scope note: malformed doubled-equals input has pre-existing chunk-dependent recovery. One malformed partition changes output under this fix; this PR does not claim uniform decoding of invalid encodings.

AI assistance was used to investigate, implement, test, and review this change.

[codspeed]

Merging this PR will not alter performance

✅ 5 untouched benchmarks
⏩ 5 skipped benchmarks¹

---

_{Comparing AryanRG:fix/quoted-printable-chunk-boundary (c738c1b) with main (51ff1a4)}

Footnotes

1. 5 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[cubic-dev-ai]

All reported issues were addressed across 2 files

<sub>Reply to a comment to ask cubic a question or push back. It learns from your replies.

View guided diff | Re-trigger cubic</sub>