[fix] Remove branching to mitigate flaky used cells

[nyunyunyunyu]

I think most flakiness(but not sure if it's all) is from nondeterministic pow_witness.
This fix cannot fully avoid flakiness. For example, here then/else branch have different numbers of instructions because then has 1 extra JAL. Therefore flakiness is natural. We can avoid distraction by not showing a small % of used cell changes.

closes INT-2323

[github-actions]

Benchmarks

group	app_log_blowup	app_total_cells_used	app_total_cycles	app_total_proof_time_ms	leaf_log_blowup	leaf_total_cells_used	leaf_total_cycles	leaf_total_proof_time_ms	instance	alloc
base64_json_program	2	(+14,572 [+0.1%]) 14,996,988	(+42 [+0.0%]) 217,352	(-35.0 [-1.4%]) 2,489.0	-	-	-	-	64cpu-linux-arm64	mimalloc
fibonacci_program	2	(-4,740 [-0.0%]) 51,512,218	1,500,219	(+17.0 [+0.3%]) 6,399.0	-	-	-	-	64cpu-linux-arm64	mimalloc
fibonacci_program	2	(-2,376 [-0.0%]) 51,512,218	1,500,219	(-454.0 [-6.3%]) 6,746.0	-	-	-	-	64cpu-linux-x64	jemalloc
regex_program	2	(+15,167 [+0.0%]) 238,565,138	(+22 [+0.0%]) 4,181,072	(-255.0 [-0.9%]) 26,923.0	-	-	-	-	64cpu-linux-arm64	mimalloc
verify_fibair	2	(-11,352 [-0.1%]) 8,294,755	(-645 [-0.3%]) 198,497	(-42.0 [-2.9%]) 1,426.0	-	-	-	-	64cpu-linux-arm64	mimalloc

Commit: 876552a

Benchmark Workflow

[linear]

INT-2323 Determine why metrics differ between CI runs when no change

[jonathanpwang]

not sure if there was flakiness before we turned on grinding, but let's see