fix: Bump CI actions and stabilize flaky notebook checks

[huseyincavusbi]

Description

Two changes:

1. Node.js deprecation warnings in CI
Bumped all GitHub Actions to latest stable versions to clear Node.js 20 deprecation warnings (checkout@v3-v4, cache@v3-v4, upload-artifact@v4-v7, download-artifact@v4-v8, setup-uv@v6-v7). Also added workflow_dispatch to checks.yml for manual runs.

2. Flaky notebook checks
Some notebooks (e.g. Santa_Coder) fail intermittently due to tiny floating-point noise across CI runs (~3.9515 vs ~3.9514). Following the same principle as the GQA allclose fix, I added a regex sanitizer to doc_sanitize.cfg that truncates floats to 3 decimal places before comparison. This preserves meaningful numerical checks while absorbing last-bit precision noise.

Type of change

• Bug fix (non-breaking change which fixes an issue)

Checklist:

• My changes generate no new warnings
• New and existing unit tests pass locally with my changes
• I have not rewritten tests relating to key interfaces which would affect backward compatibility

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[huseyincavusbi]

Thanks for pointing this out @jlarson4. I changed it to be scale aware..
What do you think about this now:

• Numbers ≥ 1 (e.g. 3.9515): truncated to 3 decimals → 3.951
• Numbers < 1 (e.g. 0.001234): truncated to 4 decimals → 0.0012

[jlarson4]

@huseyincavusbi Would it be possible have numbers < 1 match to a certain amount of significant figures, say 3 decimals? Examples:

0.001234 to 0.00123
0.01234 to 0.0123

[huseyincavusbi]

hey @jlarson4 . Updated the regex to use 3 significant figures for small numbers instead of fixed decimals. So 0.001234 now becomes 0.00123 instead of 0.001, keeps the precision where it actually matters

[jlarson4]

This looks great! I am going to merge, thank you for tackling this issue.