fix(precompile): BLS G2 MSM

[shekhirin]

• Adds back the BLS Test Vectors from AlphaNet
• Fixes bug introduced by the newer test vectors from the EIP which involved a BLST footgun in the API that took us multiple hours to identify. This behavior MUST be clearly documented as it's dangerous for broader usage of BLST.

This was a process failure from our side:

1. We shipped 2537 in AlphaNet ~1 mo ago, and thought the EIP was "done"
2. We did not look at the updated test vectors, this would've surfaced this 2-3 weeks ago

Makes me think we may want to subscribe to specific EIP updates in an internal channel to ensure we don't miss anything.

[github-actions]

Valgrind Results:

==3811== Cachegrind, a cache and branch-prediction profiler
==3811== Copyright (C) 2002-2017, and GNU GPL'd, by Nicholas Nethercote et al.
==3811== Using Valgrind-3.18.1 and LibVEX; rerun with -h for copyright info
==3811== Command: target/release/snailtracer
==3811== 
--3811-- warning: L3 cache found, using its data for the LL simulation.
Running snailtracer example!
elapsed: 1.810957493s
==3811== 
==3811== I   refs:      429,370,134
==3811== I1  misses:          3,130
==3811== LLi misses:          2,793
==3811== I1  miss rate:        0.00%
==3811== LLi miss rate:        0.00%
==3811== 
==3811== D   refs:      206,204,554  (133,650,291 rd   + 72,554,263 wr)
==3811== D1  misses:        344,322  (    204,238 rd   +    140,084 wr)
==3811== LLd misses:        137,640  (      4,278 rd   +    133,362 wr)
==3811== D1  miss rate:         0.2% (        0.2%     +        0.2%  )
==3811== LLd miss rate:         0.1% (        0.0%     +        0.2%  )
==3811== 
==3811== LL refs:           347,452  (    207,368 rd   +    140,084 wr)
==3811== LL misses:         140,433  (      7,071 rd   +    133,362 wr)
==3811== LL miss rate:          0.0% (        0.0%     +        0.2%  )

[gakonst]

Found & fixed the bug and updated PR description. Feel free to improve on my code (esp the empty slice check is ugly but it's 330am). This should be functional for Devnet-0. There might be better ways to work around this massive BLST footgun but don't have the time for this rn.

[rakita]

Pending some discussion, PR lgtm

[rakita]

lgtm