parser: fix tuple field parsing in offset_of

[Siteshcodes]

Fixes #23178.

When using offset_of with nested tuple fields such as 0.1.1.1, the lexer parses parts of the field path (0.1 and 1.1) as float literals. This makes offset_of reject the expression with a syntax error.

This PR splits float literals while parsing offset_of into field indices and dots, so the resulting events match the structure expected by OFFSET_OF_EXPR. This follows the same general approach used by rustc in break_up_float.

Also adds an inline regression test covering nested tuple fields.

AI disclosure: Used AI as an exploratory tool to navigate the parser/event plumbing and help with the refactoring.

[Siteshcodes]

The Rust, Rust Cross, rustfmt, clippy, miri, and analysis-stats checks pass.

The remaining failures are unrelated to this PR:

• proc-macro-srv (windows): failure in existing master code in crates/proc-macro-srv/src/dylib.rs
• Typo Check: existing typos in unrelated files

I reproduced the relevant failures locally and confirmed they are not introduced by this change.

[ChayimFriedman2]

I suspect you used AI without disclosing that. Please note that our AI policy requires disclosure. Make sure to read the policy and follow it.

[Siteshcodes]

Thanks for taking a look. To be fully transparent about AI usage per the policy:

Where AI helped: I used AI as an exploratory tool to navigate event.rs and syntax-bridge, trace how FloatSplitHack was originally wired up in commit 6fa6efe, and help deduplicate the float-splitting helper between FIELD_EXPR and offset_of.
What I understand and wrote: The core problem is that the lexer tokenizes 0.1 into a single FLOAT_NUMBER. In standard field access (foo.0.1), this is split into nested FIELD_EXPR nodes. offset_of needs flat NAME_REF and DOT tokens directly under OFFSET_OF_EXPR. I adapted the parsing in atom.rs, ensured the CST output matches this flat structure, and added the inline test fixture.
rustc citation: This mirrors how rustc handles float field indices in compiler/rustc_parse/src/parser/expr.rs via break_up_float().

Happy to make any adjustments or refactor further if you’d prefer a different approach here.

[ShoyuVanilla]

Tests:

* `cargo fmt --all -- --check`

* `cargo test -p parser`

* `cargo test -p syntax-bridge -p mbe`

* `git diff --check`

I guess no human would write such section to PR description

[Siteshcodes]

Fair catch — that template was indeed generated when I first opened the PR. I've updated the PR description now in my own words.