Improve base-32 hash decoding performance with reverse map

[avnik]

Motivation

This PR replaces the linear search over nix32Chars with a constexpr-initialized reverse lookup table for decoding base-32 hashes. The changes include:

• Defining nix32Chars as a constexpr char[].
• Adding a constexpr std::array<unsigned char, 256> (reverseNix32Map) to map characters to their base-32 digit values at compile time.
• Replacing the slow character search loop with a direct lookup using reverseNix32Map.
• Removing std::once_flag/isBase32 logic in references.cc in favor of reverseNix32Map

This significantly improves performance and clarity of base-32 decoding logic, while preserving correctness.

[Ericson2314]

Thanks for this! This is indeed much better: faster and clearer.

Just one small thing, then let's land it!

[avnik]

@Ericson2314 your suggestion applied

[xokdvium]

I wouldn't mind if we could add a benchmark for this, but this could be done in a follow-up.

[avnik]

@xokdvium also wouldn't be bad to add dedicated tests for both nix32 encoding/decoding

[xokdvium]

wouldn't be bad to add dedicated tests for both nix32 encoding/decoding

Absolutely. If you'd like you could handle it in a separate PR. Any help with the test coverage would be greatly appreciated.

[Ericson2314]

Note re #13682 that this PR is not supposed to affect how reference scanning works. It just affects how the parsing of hashes themselves (a Hash constructor) works, and then deduplicates the two.

[Ericson2314]

I talked to @xokdvium, and we can just do the rest in a follow-up PR.

[Ericson2314]

We are working on this now, so don't sweat @avnik.