flamenco, runtime, vm: clean up bounds on the number of instruction accounts

[topointon-jump]

Corrects the value of FD_INSTR_ACCT_MAX to a larger bound derived from the transaction MTU. Without this, it is possible to cause a conformance divergence.

Credit to @nick0ve for reporting this in https://github.com/firedancer-io/auditor-internal/issues/348.

We were also using FD_INSTR_ACCT_MAX in several places where we shouldn't have been. This PR corrects that, so that we only use this larger bound where it's actually necessary.

Thankfully, solana-foundation/solana-improvement-documents#406 fixes this by limiting the number of instruction accounts to 255 at the transaction sanitization level. Once that SIMD is implemented and activated we can use the same 255 limit everywhere.

Increases the size of fd_runtime_t by 470KB.

[github-actions]

Performance Measurements ⏳

Suite	Baseline	New	Change
backtest mainnet-368528500-perf per slot	0.050966 s	0.050701 s	-0.520% ✅
backtest mainnet-368528500-perf snapshot load	1.626 s	1.64 s	0.861% ✅
backtest mainnet-368528500-perf total elapsed	50.9661 s	50.701451 s	-0.519% ✅
firedancer mem usage with mainnet.toml	1023.23 GiB	1023.23 GiB	0.000% ✅

[github-actions]

Performance Measurements ⏳

Suite	Baseline	New	Change
backtest mainnet-368528500-perf per slot	0.083272 s	0.076843 s	-7.720% ✅
backtest mainnet-368528500-perf snapshot load	3.068 s	3.066 s	-0.065% ✅
backtest mainnet-368528500-perf total elapsed	83.271952 s	76.842749 s	-7.721% ✅
firedancer mem usage with mainnet.toml	1023.23 GiB	1023.23 GiB	0.000% ✅

[github-actions]

Performance Measurements ⏳

Suite	Baseline	New	Change
backtest mainnet-368528500-perf per slot	0.073861 s	0.073771 s	-0.122% ✅
backtest mainnet-368528500-perf snapshot load	3.18 s	3.144 s	-1.132% ✅
backtest mainnet-368528500-perf total elapsed	73.861388 s	73.770658 s	-0.123% ✅
firedancer mem usage with mainnet.toml	1023.23 GiB	1023.23 GiB	0.000% ✅

[github-actions]

Performance Measurements ⏳

Suite	Baseline	New	Change
backtest mainnet-368528500-perf per slot	0.062897 s	0.063242 s	0.549% ✅
backtest mainnet-368528500-perf snapshot load	2.345 s	2.343 s	-0.085% ✅
backtest mainnet-368528500-perf total elapsed	62.897126 s	63.242102 s	0.548% ✅
firedancer mem usage with mainnet.toml	1023.23 GiB	1023.23 GiB	0.000% ✅

[github-actions]

Performance Measurements ⏳

Suite	Baseline	New	Change
backtest mainnet-368528500-perf per slot	0.073589 s	0.074041 s	0.614% ✅
backtest mainnet-368528500-perf snapshot load	3.208 s	3.155 s	-1.652% ✅
backtest mainnet-368528500-perf total elapsed	73.589184 s	74.041384 s	0.614% ✅
firedancer mem usage with mainnet.toml	1023.23 GiB	1023.23 GiB	0.000% ✅

[github-actions]

Performance Measurements ⏳

Suite	Baseline	New	Change
backtest mainnet-368528500-perf per slot	0.050908 s	0.050993 s	0.167% ✅
backtest mainnet-368528500-perf snapshot load	1.612 s	1.609 s	-0.186% ✅
backtest mainnet-368528500-perf total elapsed	50.90766 s	50.992866 s	0.167% ✅
firedancer mem usage with mainnet.toml	1023.23 GiB	1023.23 GiB	0.000% ✅

[github-actions]

Performance Measurements ⏳

Suite	Baseline	New	Change
backtest mainnet-368528500-perf per slot	0.073527 s	0.07412 s	0.807% ✅
backtest mainnet-368528500-perf snapshot load	3.194 s	3.206 s	0.376% ✅
backtest mainnet-368528500-perf total elapsed	73.527075 s	74.12003 s	0.806% ✅
firedancer mem usage with mainnet.toml	1023.23 GiB	1023.23 GiB	0.000% ✅

[ibhatt-jumptrading]

comments shouldn't exceed 80 chars

[ripatel-fd]

The fix looks good, but can we add a manual test passing a worst-case transaction into the runtime? This makes it easier to verify the behavior in the future without having to rely on external fuzzing infra.