Eagerly perform set insertions in nub{Ord,Int}

[meooow25]

This results in an improvement in overall performance. See the added
Note [Eager set insertions] for an explanation.

Also mark the go functions as strict in the input set.

Benchmarks on GHC 9.14 show a decrease in allocations of 8% on
nubOrd.noFusion_distinct and nubOrd.strings_noFusion_distinct.

---

Name                                Time - - - - - - - -    Allocated - - - - -
                                         A       B     %         A       B     %
nubOrd.issue1202_distinct            46 μs   42 μs   -8%    672 KB  618 KB   -8%
nubOrd.issue1202_repeat             4.0 μs  3.9 μs   -2%     16 KB   16 KB   +0%
nubOrd.noFusion_distinct             44 μs   39 μs   -9%    633 KB  579 KB   -8%
nubOrd.noFusion_repeat              1.8 μs  1.3 μs  -28%    152 B    96 B   -36%
nubOrd.strings_noFusion_distinct    114 μs  112 μs   -1%    626 KB  572 KB   -8%
nubOrd.strings_noFusion_repeat      4.2 μs  4.0 μs   -4%    154 B    98 B   -36%

[meooow25]

Note for the record: The surprisingly large time improvement in nubOrd.noFusion_repeat compared to nubOrd.strings_noFusion_repeat is because, for some reason, this change convinces GHC to inline memberIndex only in nubOrd.noFusion_repeat.

I may look into this afterwards, but for now with this PR I just want to avoid the thunks.
It looks like -fworker-wrapper-cbv also helps with these benchmarks (#827).