Optimize `hasAttribute`

[sim642]

It's currently implemented as

cil/src/cil.ml

Lines 1429 to 1430 in bf39f60

	let hasAttribute s al =
	(filterAttributes s al <> [])

This is silly because filterAttributes will allocate a new list, just to check containment. This should be done with List.exists or something reasonable that doesn't allocate at all.

This showed up as a notable source of allocations while analyzing rsync. The hasAttribute is used by bitsOffset calculation (to look for packing attributes) which we heavily use in Goblint during semantic_equal of addresses (and their sets).

In addition to the one-liner change, we should evaluate the impact on rsync/sv-benchmarks.

Besides fixing this, it might make sense to also cache the bitsOffset calls on Goblint side.

[michael-schwarz]

It's really remarkable which places turn out to be hot when properly benchmarking. I think analyze that is worth it just for these insights already.

[sim642]

This should be done with List.exists or something reasonable that doesn't allocate at all.

I might've been overly optimistic: List.exists will still cause the predicate closure to be allocated but that also already was happening in List.filter.
So it would save the list allocation and short-circuit but perhaps not completely eliminate the overhead.

Thus, goblint/analyzer#1964 might avoid the overhead further.