[fix] expr builder eval inconsistency

[luffykai]

Issue: the way we do evaluate_overflow_isize and evaluate_overflow_expr were inconsistent for Select
Since we had flag available for isize we directly using it to choose left or right, but that's not viable in expr. But this inconsistency MIGHT result in different max_overflow_bit in some condition (=> ended up in range checker error), which we only discovered with BLS curve double setup...

[jonathanpwang]

how is this different than when you just use dbg!?

[luffykai]

when adding print to mod-builder, it applies to all chips using it. But usually when debugging i only care about one chip. So the way I use this is to call set_debug only at the specific chip

[jonathanpwang]

LGTM but can you move the test to pairing tests crate instead

[github-actions]

group	app.proof_time_ms	app.cycles	app.cells_used	leaf.proof_time_ms	leaf.cycles	leaf.cells_used
verify_fibair	(+26 [+1.2%]) 2,207	513,827	19,376,791	-	-	-
fibonacci_program	(+397 [+7.7%]) 5,541	1,500,137	51,487,838	-	-	-
regex_program	(+3066 [+21.0%]) 17,664	4,190,904	165,010,909	-	-	-
ecrecover_program	(-94 [-3.6%]) 2,494	285,401	15,075,033	-	-	-

Commit: d507515

Benchmark Workflow