perf(precompile): zero-copy scalar conversion in BLS12-381 MSM

[dizer-ti]

Avoid allocating a temporary Vec when converting scalars to bytes in p1_msm/p2_msm.
blst_scalar is repr(C) with a single [u8; 32] field, so we can safely reinterpret the slice.

[codspeed-hq]

Merging this PR will improve performance by 3.32%

⚡ 1 improved benchmark
✅ 172 untouched benchmarks

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	transfer	14.5 µs	14.1 µs	+3.32%

---

_{Comparing dizer-ti:perf/bls-msm-zero-copy-scalars (e8021c1) with main (087ac46)}

[rakita]

lgtm