Optimize parsing and serialization performance

[andrew]

Replace regex validators with hand-written character checks, use strings.Builder for ToString, implement custom purl parser, and avoid reflection-based sorting. Also fixes query string escaping for qualifiers and subpaths per purl spec.

Spec submodule updated and all tests passing.

Benchmarks on Apple M1 Pro:

Parsing (FromString)

Benchmark	Before	After	Speedup	Memory
Simple	335 ns	155 ns	2.2x	184B → 40B
Namespace	412 ns	191 ns	2.2x	184B → 40B
Complex	1811 ns	551 ns	3.3x	584B → 360B

Serialization (ToString)

Benchmark	Before	After	Speedup	Memory
Simple	192 ns	61 ns	3.2x	136B/5 allocs → 64B/1 alloc
Namespace	291 ns	118 ns	2.5x	248B/6 allocs → 96B/1 alloc
Complex	897 ns	273 ns	3.3x	776B/21 allocs → 336B/3 allocs

Validation (regex → hand-written)

Benchmark	Before	After	Speedup
ValidQualifierKey	669 ns	32 ns	21x
ValidType	559 ns	22 ns	25x

Roundtrip (Parse + Serialize)

Benchmark	Before	After	Speedup	Memory
Simple	550 ns	233 ns	2.4x	320B → 104B
Complex	2978 ns	858 ns	3.5x	1360B → 696B

[petergardfjall]

Thanks for the PR! There is a rather big PR in the pipeline (#83) that attempts to adapt the code to pass all the new tests in the upstream purl-spec. It is awaiting a second round of approval. I'd prefer having that merged first before applying optimizations.

[andrew]

I have also fixed almost all of the tests in this branch

[andrew]

@petergardfjall I've updated the spec submodule and got all tests passing, so this should be good to go. I'm now shipping it in https://github.com/git-pkgs/git-pkgs because the current version has some real slow bits when parsing thousands of purls.