Bump typelit plugins (preparations)

[kleinreact]

The PR introduces some preparations for #3061 to make Clash.Sized.Vector compile with ghc-typelits-natnormalise-0.9.1, ghc-typelits-knownnat-0.8.2, and ghc-typelits-extra-0.5.1.

The update avoids the erroneous, but currently required unification under type families (cf. clash-lang/ghc-typelits-natnormalise#108), which has been fixed by clash-lang/ghc-typelits-natnormalise@c994419.

The modifications are only introduced for GHCs between 9.4 and 9.10, as the introduced fix won't work for GHCs prior to 9.4 and is not needed any more with GHC 9.12 due to ghc/ghc@a63ee33.

Still TODO:

• Write a changelog entry (see changelog/README.md)
• Check copyright notices are up to date in edited files

[kleinreact]

I have now clue why the doctest is suddenly failing now.

In terms of semantics the implementation should not have changed in any way at all.

<no location info>: warning: [GHC-98887]
    -XGeneralizedNewtypeDeriving is not allowed in Safe Haskell; ignoring -XGeneralizedNewtypeDeriving
[ERROR  ] [ThreadId 638] /builds/clash-lang/clash-compiler/clash-prelude/src/Clash/Sized/RTree.hs:531: failure in expression `replaceTree 9 6 (BR (BR (LR 1) (LR 2)) (BR (LR 3) (LR 4)))'
[ERROR  ] [ThreadId 638] expected: <<1,2>,<3,*** Exception: Clash.Sized.Vector.replace: index 9 is out of bounds: [0..3]
[ERROR  ] [ThreadId 638]           ...
[ERROR  ] [ThreadId 638]  but got: <<1,2>,<3,4>>
[ERROR  ] [ThreadId 638]                     ^

[christiaanb]

The old version of go was strict in the Vec argument.

[kleinreact]

I added the strictness requirement in the argument again, which fixes the failing doctest. However, the pattern matches require some additional inaccessibility proofs for the non-reachable cases then, which have been added as well.

This version will also work with the updated plugins avoiding the issue reported in clash-lang/ghc-typelits-natnormalise#108.