Consider whether native bignum backend should be used by default

In the past week have have three distinct tickets which were traceable back to gmp's lack of releases: one undefined behavior issue (#22497 (closed)), one exploitable buffer overflow (#22789 (closed)), and one WebAssembly-specific stack overflow (#22602 (closed)). This, coupled with the fact that gmp adds complexity to our packaging story, makes me wonder whether we wouldn't be better off simply switching our binary distributions to use the Haskell-native bignum backend by default. Last I recall, the native backend implements most common operations reasonably efficiency and bignum operations constitute a vanishingly small fraction of CPU time in most programs.