Meter ignoring filters

[etki]

Yet another performance PR, this is probably the most beneficial one among all, avoiding extra stream overhead, zero-sized list, O(n²). It could probably get a whole order of a magnitude speed up if #6113 would be implemented (we can then create a same-sized array and fill it in a single iteration, then create tags without any unnecessary re-sorting).

I'm having too many benchmark reports in my files to be sure i'm picking the right one, so i'll rerun them soon. Just to get a glimpse, this is what i'm looking at and what could be not the latest version (the last ones would be pretty rare in real applications; they would also usually filter around 62 tags out of 64, thus the drop compared to 8/32 combination). Set.of(), available only for java 9+, would be even more performant if supplied by end users.

ignored tags	meter tags	version	ns/op	instructions
8	32	main	2793.114 ± 2.144	12508.501
8	32	PR	1595.240 ± 1.976	5541.988
64	64	main	8439.085 ± 7.119	38896.080
64	64	PR	1161.565 ± 1.578	4052.474