[pylint] Implement pointless-string-statement (PLW0105) - #27253

Summary

• I noticed that ruff had not yet ported over pylint's pointless-string-statement (W0105) rule, after finding some offenses in our own codebase and figuring there would be enough source material to make the fix upstream
• accordingly, this checks off one parity box from Implement Pylint #970, by implementing pointless-string-statement as PLW0105
• I based the implementation off of pylint (rather than PEP 257) for parity, and it's worth noting that they differ in one spot: PEP 257 also recognizes "additional docstrings" (a string right after the real docstring), but they are flagged by pylint. No strong opinion here but I kept the pylint logic for parity and documented the difference in code; happy to change if desired.

Changes

• dispatched from analyze/suite.rs (same entry point as PYI021) because the exemptions need the previous sibling statement
• didn't use the existing ExpectedDocstringKind::Attribute since it was narrower than pylint's exemption (no __init__, no TypeAlias, single-Name targets only)
• ended up having to touch flake8-bugbear in order to generalize at_last_top_level_expression_in_cell for the sake of notebook handling: strings that are a cell's last top-level expression are the cell's displayed output and are exempted, matching B015/B018. The rule runs at suite-analysis time, where current_statement() isn't populated, so the helper now takes the statement's end offset as a parameter instead. Alternatives to generalizing in-place are duplicating some logic into the pylint rule, or relocating the helper. (Note that B015/B018 behavior is unchanged and their tests were not modified.)
• no remediation path here because the fix depends on the original intent

Test Plan

• ported over pylint's test coverage for the rule, and added new notebook tests
• compared to a run of pylint against a large private codebase; matched line-for-line on 93 offenses