Improve fromList for IntSet and IntMap

[meooow25]

Implement a fusion-friendly version of the smarter algorithm proposed in #653.

For #1128.

---

Benchmarks on GHC 9.10.1:

IntSet:

Name                          Time - - - - - - - -    Allocated - - - - -
                                   A       B     %         A       B     %
fromList:asc                   46 μs   18 μs  -60%    480 KB  5.5 KB  -98%
fromList:asc:fusion            39 μs  4.1 μs  -89%    480 KB  5.5 KB  -98%
fromList:asc:sparse            95 μs   47 μs  -50%    864 KB  352 KB  -59%
fromList:asc:sparse:fusion     87 μs   41 μs  -52%    864 KB  352 KB  -59%
fromList:random               459 μs  467 μs   +1%    1.5 MB  2.4 MB  +67%
fromList:random:fusion        450 μs  459 μs   +1%    1.5 MB  2.4 MB  +67%
fromList:randomDups            57 μs   48 μs  -16%    352 KB  300 KB  -14%
fromList:randomDups:fusion     52 μs   48 μs   -7%    352 KB  300 KB  -14%
fromList:worstCase            711 μs  824 μs  +15%    6.9 MB   14 MB  +96%

IntMap:

Name                              Time - - - - - - - -    Allocated - - - - -
                                       A       B     %         A       B     %
fromList:asc                      100 μs   48 μs  -52%    864 KB  352 KB  -59%
fromList:asc:fusion                88 μs   37 μs  -57%    864 KB  352 KB  -59%
fromList:random                   480 μs  463 μs   -3%    1.5 MB  2.4 MB  +67%
fromList:random:fusion            452 μs  448 μs   +0%    1.5 MB  2.4 MB  +67%
fromList:randomDups               309 μs  263 μs  -15%    1.1 MB  1.7 MB  +57%
fromList:randomDups:fusion        296 μs  247 μs  -16%    1.1 MB  1.7 MB  +57%
fromList:worstCase                982 μs  1.1 ms  +16%    7.3 MB   14 MB  +97%
fromListWith:randomDups           316 μs  302 μs   -4%    1.3 MB  1.9 MB  +54%
fromListWith:randomDups:fusion    321 μs  335 μs   +4%    1.3 MB  2.3 MB  +86%

[meooow25]

The worse performance of fromListWith:randomDups:fusion compared to fromListWith:randomDups is due to a fusion issue: GHC #25259. This previously showed up in #1042 too. Since we have a potential fix, i.e. defining our old foldl', we can avoid the problem, but I think it's better to just get this fixed in base.

[meooow25]

I'm working on getting this fixed in base, but making moveToB NOINLINE also seems to do the trick, making the function small enough to not run into GHC #25259. This is good anyway because we are reducing the amount of code we inline.