[blouirts123]

Raw strings currently retain both backslashes in an escaped backslash, so a filter looking for a Windows path segment ending in \ can match the wrong rows. Decode the two raw-string escape forms in one pass, preserving unknown escapes such as \z and \u.

This follows the current specification's raw-string grammar. The existing compliance case is updated to that grammar; the older JEP-12 page uses different wording. Related to #323.

Validation: full suites on Python 3.12 and 3.14: 995 passed, 1 skipped; independent compliance/focused run: 924 passed; root focused tests: 4 passed. Existing deprecation warnings remain.

Adjacent #349 adds lexer unit coverage without implementing this decoding; #382 fixes token offsets, which this change leaves untouched.

[carlitogwapito]

Small companion for adjacent escapes: 20-line test-only commit, based on this branch with no implementation changes.

Two useful guards: an escaped backslash beside an escaped quote, and an escaped backslash beside a preserved \u1234 sequence. The latter should remain literal rather than becoming a Unicode character or losing another backslash. These compose the two escape forms in the current grammar, instead of testing them only in isolation.

Both new test methods pass against this branch's complete package source on Windows/Python 3.14.4, checking lexer values and public search. The base fails both at the lexer assertion, and a separate check confirms search returns the wrong value too. The existing unknown-escape and escaped-quote controls pass on both sources, with fresh package/parser state for each case.

Focused Ruff lint/format checks pass. I haven't run the full suite or version matrix, and these don't address the separate token-offset work. With the file added to this branch, the command is python -m unittest tests.test_raw_string_escape_composition. Take whichever cases help.