perf: lock-free cycle count increment

[jonathanpwang]

A stupid but important performance optimization when metric collection is turned on: every instruction increments the cycle count. When this is done directly via metric::Counter the metric recorder acquires a lock every time. This has significant (2x) performance impact on execution.

[github-actions]

group	app.proof_time_ms	app.cycles	app.cells_used	leaf.proof_time_ms	leaf.cycles	leaf.cells_used
verify_fibair	(-105 [-6.5%]) 1,503	194,752	8,012,822	-	-	-
fibonacci_program	(-858 [-11.6%]) 6,556	1,500,137	51,505,102	(-1830 [-10.2%]) 16,194	3,171,679	128,849,226
regex_program	(-2749 [-12.3%]) 19,623	4,190,904	165,028,173	(-4083 [-11.5%]) 31,546	6,523,215	291,291,355
ecrecover_program	(-190 [-6.5%]) 2,732	290,016	15,230,037	(-6759 [-13.9%]) 41,976	9,778,270	445,631,858

Commit: 2bbc8fc

Benchmark Workflow