perf: enable x86-64-v2 kernels on a native v2 baseline

[anonrig]

Summary

Addresses #640: Windows 11 24H2, Rocky Linux 9, and RHEL 9 default to x86-64-v2 (SSSE3, SSE4.1, SSE4.2, POPCNT). Earlier revisions forced those kernels into default gcc/clang x86-64 builds with target("ssse3"). That changed the unity ada.cpp inlining budget on CodSpeed (which compiles the default x86-64 baseline, not v2): SetHash went from −12% to −15% with in-TU wrappers, and was still −7% after moving the kernels to a separate TU.

This revision does not pretend the default gcc TU is SSSE3. Rocky/RHEL 9 already define __SSSE3__ and compile the existing has_tabs_or_newline / host-delimiter SSSE3 paths. On that native v2 baseline (and on MSVC /arch:SSE4.2 or AVX):

• Authority-delimiter scans (@ / \ ?) use an SSSE3 pshufb nibble table, matching the host-delimiter kernels already on main.
• has_tabs_or_newline uses SSE4.1 ptest on the common miss path.
• ADA_X86_64_V2 applies -march=x86-64-v2 (MSVC /arch:SSE4.2) when a distributor wants that baseline explicitly.

Default gcc/clang x86-64 codegen for setters matches main (SSE2 has_tabs, scalar authority tables). Parser / percent-encode keep target("ssse3") as on main. Public API and ABI are unchanged.

Test plan

• Public-API tests for long-host parse, long userinfo, IPv6 [::1]:8080, tabs/newlines in a 16-byte window (including overlapping tail), and set_hash
• Default gcc (ADA_X86_64_V2 off): preprocessor has ADA_SSE2 only; 351/351 tests passed (demo/cdemo/fuzzer/singleheader skipped locally)
• -DADA_X86_64_V2=ON: preprocessor has ADA_SSSE3/ADA_SSE41/ADA_SSE42/ADA_POPCNT; 351/351 tests passed
• CI green: all 55 checks on de728283, including lint-and-format, ARM/macOS/MSVC/clang-cl, ABI, sanitizers, fuzzing
• CodSpeed: not alter performance vs main (47 untouched; SetHash / SetProtocol no longer regress)

Closes #640

  

[codspeed-hq]

Merging this PR will not alter performance

✅ 47 untouched benchmarks
⏩ 4 skipped benchmarks¹

---

_{Comparing cursor/x86-64-v2-optimizations-f0f8 (de72828) with main (e675447)}

Footnotes

1. 4 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 63.27%. Comparing base (e675447) to head (de72828).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1243   +/-   ##
=======================================
  Coverage   63.27%   63.27%           
=======================================
  Files          39       39           
  Lines        7705     7705           
  Branches     3514     3514           
=======================================
  Hits         4875     4875           
  Misses        752      752           
  Partials     2078     2078           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[lemire]

@anonrig I think that this should be tested. The x64 processors on GH are surely v2 (it is hard or impossible to find processors than are not v2), so testing this flag in CI should work.