feat: implement PubkeyIndexMap

[twoeths]

Description

• implement PubkeyIndexMap in zig
• write C-ABI functions and build shared library
• write Bun binding code, by loading the shared library
• unit tests + CI + benchmark

[github-actions]

Performance Report

✔️ no performance regression detected

Full benchmark results

Benchmark suite	Current: 6551d02	Previous: null	Ratio
get values - 1000	406.00 ns/op
get values - naive - 1000	617.00 ns/op
set values - 1000	416.00 ns/op
set values - naive - 1000	677.00 ns/op
get values - 1000000	1.3880 us/op
get values - naive - 1000000	1.8760 us/op
set values - 1000000	1.0490 us/op
set values - naive - 1000000	2.6510 us/op

by benchmarkbot/action

[twoeths]

I run this branch on my laptop, got

test/perf/index.test.ts
  get/set
    ✔ get values - 1000                                                    7874016 ops/s    127.0000 ns/op        -    4142957 runs   1.32 s
    ✔ get values - naive - 1000                                            6060606 ops/s    165.0000 ns/op        -    3678292 runs   1.31 s
    ✔ set values - 1000                                                    7633588 ops/s    131.0000 ns/op        -    7731718 runs   2.44 s
    ✔ set values - naive - 1000                                            3690037 ops/s    271.0000 ns/op        -    1349665 runs  0.606 s
    ✔ get values - 1000000                                                 1564945 ops/s    639.0000 ns/op        -    1358041 runs   1.11 s
    ✔ get values - naive - 1000000                                         1048218 ops/s    954.0000 ns/op        -     902932 runs   1.01 s
    ✔ set values - 1000000                                                 1584786 ops/s    631.0000 ns/op        -     973566 runs  0.808 s
    ✔ set values - naive - 1000000                                        700770.8 ops/s    1.427000 us/op        -     247581 runs  0.404 s

which is faster than napi-rs (blst-ts)

get/set
    ✓ get values - 1000                                                    4115226 ops/s    243.0000 ns/op   x1.004    1633842 runs  0.606 s
    ✓ get values - naive - 1000                                            5208333 ops/s    192.0000 ns/op   x0.950    1935707 runs  0.606 s
    ✓ set values - 1000                                                    4016064 ops/s    249.0000 ns/op   x1.016    1070166 runs  0.404 s
    ✓ set values - naive - 1000                                            2898551 ops/s    345.0000 ns/op   x0.953    1304631 runs  0.606 s
    ✓ get values - 1000000                                                 1298701 ops/s    770.0000 ns/op   x0.905    1012602 runs  0.909 s
    ✓ get values - naive - 1000000                                         1658375 ops/s    603.0000 ns/op   x0.907     701857 runs  0.505 s
    ✓ set values - 1000000                                                 1287001 ops/s    777.0000 ns/op   x0.898     668077 runs  0.606 s
    ✓ set values - naive - 1000000                                        781250.0 ops/s    1.280000 us/op   x0.948     864027 runs   1.21 s

(note that the naive is not the same due to different benchmark framework used)

[twoeths]

note that I tried to build a HashMap with []u8 as key to avoid redundant @memcpy but it's even slower
persisted the work in te/pubkey_index_map_no_copy branch for later reference

// the same to std.AutoHashMap except for the hash() and eql() functions
const U8SliceHashMap = std.HashMap(Key, u32, U8SliceContext, default_max_load_percentage);
const hash_fn = std.hash_map.getAutoHashFn(Key, U8SliceContext);
const U8SliceContext = struct {
    pub fn hash(_: U8SliceContext, key: Key) u64 {
        return std.hash.Wyhash.hash(0, key);
    }

    pub fn eql(_: U8SliceContext, a: Key, b: Key) bool {
        return std.mem.eql(u8, a, b);
    }
};