chore: precomputed circuits

[ledwards2225]

.

[ledwards2225]

All flows are ecdsar1+<flow>+sponsored_fpc. Dedicated bench machine, 4 runs each with the first discarded as warmup, median of the remaining 3 reported.
Baseline = NO_PIPELINE=1 NO_TRIM=1; pipelined = NO_TRIM=1. Peak memory is /usr/bin/time -f '%M' (getrusage().ru_maxrss, same metric the CI
dashboard publishes via memusage).

Flow	Wall baseline	Wall pipelined	Wall delta	Peak baseline	Peak pipelined	Peak delta
transfer_0_recursions	5.69 s	5.62 s	-1.2%	354 MB	417 MB	+17.8%
amm_add_liquidity_1_recursions	11.97 s	11.41 s	-4.7%	734 MB	838 MB	+14.2%
storage_proof_7_layers	16.06 s	15.10 s	-6.0%	1112 MB	1467 MB	+31.9%

[ledwards2225]

Same comparisons for AMM flow with one call to malloc_trim per accumulate call:

Mode	Wall med	Wall range	Peak med	Peak range	Wall delta	Peak delta
baseline	11.95 s	11.93 - 11.98	756 MB	734 - 757	-	-
pipelined	11.45 s	11.44 - 11.48	863 MB	822 - 882	-0.50 s (-4.2%)	+107 MB (+14.2%)
baseline-trim	12.77 s	12.76 - 12.79	591 MB	575 - 597	+0.82 s (+6.9%)	-165 MB (-21.8%)
pipelined-trim	12.27 s	12.24 - 12.28	605 MB	592 - 610	+0.32 s (+2.7%)	-151 MB (-20.0%)