Match generated Python 3.15 signature defaults and names - #9795

Summary

Split the generator-supported corrections out of #9756, as requested in #9756 (comment). This PR is based directly on main and does not depend on the deferred override work.

• Match symbolic/default spelling in array, blake2, codecs, faulthandler, marshal, select, warnings, and IO seek.
• Correct _imp.create_dynamic and struct-sequence parameter names.
• Match the Future/Task add_done_callback context default. Preserve the runtime distinction: omission copies the registration context, while explicit None reaches call_soon and uses the completion context.
• Correct _csv.register_dialect's invalid-name error text, retaining name validation before dialect validation.

The grouped str/bytes/bytearray signatures and CSV display overrides remain in #9756. There is no py_optional_group, DefaultRepr::OptionalGroup, or display-only FromArgs metadata in this PR.

Validation

At final head 113460dd9:

• Configured commit hooks passed.
• Workspace Rust tests and doctests passed (1,540 passed, 20 ignored across their reported groups).
• C-API tests passed (149 passed, 2 ignored).
• Strict workspace/all-target and separate C-API Clippy passed.
• Release build passed.
• All three affected signature/asyncio/CSV snippets passed on RustPython and CPython 3.15.1; the three selected RustPython pytest cases passed.
• Standard-library regression run passed: test_array test_structseq test_hashlib test_select test_codecs test_warnings test_marshal test_faulthandler test_inspect test_csv test_io test_asyncio.test_futures test_asyncio.test_context, covering 21 files, 3,573 tests run and 225 skipped.
• Both split branches merge cleanly and preserve all original Python test statements from Add checked legacy Python 3.15 text signature overrides #9756 across the two PRs.

Local validation used a clean rebuild of every workspace package after discarding stale cross-worktree artifacts. No source fix or extra push was needed. GitHub CI is tracked separately; its final results are not claimed here.

AI assistance

Split, implementation review, and validation were performed with Codex. This reuses the generator-supported changes from #9756 and preserves their regression tests.