Use string concat instead of format

[gzuidhof]

I believe this is faster.. but who knows, benchmarks on my device are all over the place as my environment isn't consistent at all 😅 - so do give it a run yourself!

Using string concat (new)

% just benchmark --skip-single -d 20s

--------------------------------------------------------------------
MULTI-THREAD BENCHMARK (8 threads)
--------------------------------------------------------------------

Test         reqs/sec    % baseline   avg (ms)   p95 (ms)   p99 (ms)
--------------------------------------------------------------------
Baseline        24958           100       0.36       0.85       3.31
Lua (noop)      23560            94       0.44       1.24       5.43
JA4             22230            89       0.67       1.94       8.51

Using string formatting (old)

% just benchmark --skip-single -d 20s

--------------------------------------------------------------------
MULTI-THREAD BENCHMARK (8 threads)
--------------------------------------------------------------------

Test         reqs/sec    % baseline   avg (ms)   p95 (ms)   p99 (ms)
--------------------------------------------------------------------
Baseline        27374           100       0.36       0.73       3.33
Lua (noop)      23661            86       1.15       2.67      14.82
JA4             20345            74       0.42       0.85       3.90

[gzuidhof]

Note how far apart my baselines are.. yeah..

[jamielinux]

Unless I'm going crazy, I don't think just >= 2.0.0 exists?

[jamielinux]

Ah, I spotted I'm using a feature that was introduced in 1.46.0.

[jamielinux]

-- String concatenation is faster than string.format in LuaJIT

Haproxy doesn't use LuaJIT, but string concat should still be faster than string.format with normal Lua because multiple .. get compiled into a single OP_CONCAT instruction that just sums lengths and copies, skipping the per-specifier format string parsing that string.format does.

fingerprint_prefix is only once-per-request (whereas the existing hex lookup table is many-per-request). So there might be a small performance benefit to using string concat but I imagine it's probably <1%.

Still, I think the code looks nicer with string concat instead of string.format! I'll close this PR and implement a similar string concat but without type coercion and using fewer concat operands.