Speed up Balsa HTTP/1 token validation

[benni-rogge]

Commit Message:
Speed up Balsa HTTP/1 token validation

Additional Description:
Replace the per-character binary search used for Balsa HTTP/1 token validation with a constexpr bit mask. Header field names and custom methods still use the RFC 9110 token character set, but the hot path now does a fixed-index mask lookup for each byte.

Testing:

• bazel test -c opt //test/common/http/http1:balsa_parser_test //test/common/http/http1:balsa_parser_benchmark_test //test/common/http/http1:codec_impl_test
• git diff --check

Benchmark:
Command used for both baseline and optimized runs:
bazel run -c opt //test/common/http/http1:balsa_parser_benchmark -- --benchmark_filter=bmParseHeaders --benchmark_min_time=0.25s --benchmark_repetitions=10 --benchmark_out=<file>.json --benchmark_out_format=json

Shape	Headers	Baseline mean	Optimized mean	CPU reduction	95% CI
long	8	1724.6 ns	1051.0 ns	39.06%	37.59%..40.52%
long	16	3238.2 ns	1832.0 ns	43.42%	42.81%..44.03%
long	64	12156.5 ns	6246.1 ns	48.62%	47.04%..50.20%
long	256	47039.9 ns	24059.3 ns	48.85%	48.25%..49.46%
long	512	97201.0 ns	48013.2 ns	50.60%	49.46%..51.75%
short	8	531.1 ns	497.9 ns	6.26%	5.38%..7.14%
short	16	850.9 ns	744.9 ns	12.46%	11.61%..13.31%
short	64	2540.0 ns	2066.9 ns	18.62%	17.76%..19.48%
short	256	9711.9 ns	7381.7 ns	23.99%	23.50%..24.49%
short	512	19393.7 ns	14645.0 ns	24.49%	23.91%..25.06%
full-token-set	8	4054.1 ns	1984.0 ns	51.06%	49.85%..52.27%
full-token-set	16	7796.9 ns	3646.6 ns	53.23%	52.41%..54.05%
full-token-set	64	29938.3 ns	13281.4 ns	55.64%	55.13%..56.15%
full-token-set	256	120034.2 ns	51514.2 ns	57.08%	56.45%..57.72%
full-token-set	512	240048.1 ns	102724.5 ns	57.21%	56.72%..57.70%

[repokitteh-read-only]

Hi @benni-rogge, welcome and thank you for your contribution.

We will try to review your Pull Request as quickly as possible.

In the meantime, please take a look at the contribution guidelines if you have not done so already.

🐱

Caused by: #45186 was opened by benni-rogge.

see: more, trace.