Add _value API for number literals in proc-macro

[GuillaumeGomez]

Part of #136652.

This PR adds the *_value for numbers (integers and floats). However, f16 and f128 are voluntarily left out as they're still unstable. Adding support for them is just a matter of uncommenting two lines, so should be fine.

Setting same reviewer as last time.

r? @Urgau

[GuillaumeGomez]

My favourite kind of failure: works locally but not in CI. T_T

Updated the value in the tests to see which one exactly is failing.

[GuillaumeGomez]

Bug comes from an issue with GCC proc-macro ABI. Ignoring this test on GCC backend for now.

[GuillaumeGomez]

CI fixed. :)

[Urgau]

A libs (or maybe even libs-api) reviewer seems more appropriate to review this PR.

r? Amanieu (since you reviewed #136355)

[Urgau]

I missed it already has an accepted ACP. The PR can be reviewed by libs reviewer.

r? libs

[Mark-Simulacrum]

Is this exactly in sync with what the rustc parser permits?

I guess that would be a mix of https://doc.rust-lang.org/nightly/reference/tokens.html#grammar-INTEGER_LITERAL and whatever we then tokenize from it?

Perhaps this should go over the bridge and ask rustc to parse the string for us?

[GuillaumeGomez]

There is no code in rustc currently which would allow this conversion without including a lot of other things. My code is a simplified version of integer_lit (from rustc_ast/src/util/literal.rs) which parses both the number, its base and its suffix (which we already have) and returns a LitKind, which is also unnecessary here since we already know the end type.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[GuillaumeGomez]

Added the method for f16, renamed the function only_digits into strip_underscores and improve the code of parse_number a bit to remove allocations.