[Lysxia]

After messing with the printing of integers in #716, I noticed that the parsing is naive and quadratic. This implements a quasi-linear algorithm by using the same divide-and-conquer technique as for printing.

For the lazy version, I make use of splitAtWord, so I moved it and export it from an internal module. While I was at it, I removed the awkward import of a custom strict pair, instead directly ensuring that the result doesn't produce thunks.

Added some very basic benchmarks for comparison with read/readHex on strings, mainly as a sanity check, but we can observe that we are quite faster even with a lazy text made of single-character chunks.

All
  ReadBigNumber
    Decimal
      StrictText: OK
        12.5 μs ± 1.0 μs
      LazyText:   OK
        47.2 μs ± 909 ns
      String:     OK
        64.7 μs ± 5.4 μs
    Hex
      StrictText: OK
        14.0 μs ± 256 ns
      LazyText:   OK
        49.7 μs ± 2.9 μs
      String:     OK
        80.7 μs ± 4.3 μs