* Bump CI to GHC 9.12.2 and ubuntu-24.04 (needed)

* Relax bounds for GHC 9.12 (needed)

The property tested is covered by the test case above, but
the property claimed in the test name is plain wrong:

   m1 ⊈ m2  ⇒  m1 ∪ m2 ⊈ m1:          FAIL
     *** Failed! Falsified (after 1 test and 2 shrinks):
     fromList [(K {hash = 0, _x = A},0)]
     fromList []

…tor of 32 (#494)

Fixes #493

…rms (#501)

The internal `submapBitmapIndexed` function used by these functions could enter an infinite loop while comparing two nodes `m1` and `m2` where `m2` contained a sub-node associated with the partial hash `0b11111`. In this case the high bit of the combined bitmap of `m1` and `m2`, `b1Orb2`, was `1`. After checking the submap condition at the high bit, the 32-bit `m` variable used to iterate over `b1Orb2` would be left-shifted and overflow to `0`, resulting in the infinite loop. 

https://github.com/haskell-unordered-containers/unordered-containers/blob/37eee2290cab287c0947d1de20235a37ced63c94/Data/HashMap/Internal.hs#L1551-L1572

To fix this bug and any similar issues related to a branching factor of 32, we return to a branching factor of 16 for 32-bit platforms. On 64-bit platforms the branching factors stays at 32.

Fixes #491.

Mirroring haskell/containers#1143

Fixes the incorrect documentation in Data.HashMap.Lazy.fromList
/ Data.HashMap.Internal.fromList: This takes linearithmic time
rather than linear time, since we do n insertions, each of which
takes (assuming no collisions!) log32(n) steps as per the documentation
of insert.

Fixes #309

Resolves #487.

...and improve the disclaimer for D.HM.toList.

Fixes #481.