bug: restrict nbDigits in `std/math/bitslice` `Partition` method

[ivokub]

Description

The method was provided for allowing to efficiently split variables into lower and upper part without needing for a binary decomposition. It works well when nbDigits is smaller than the modulus bitlength, but in case nbDigits equal modulus bitlength (or is undefined), then we're currently underconstraining the result as we could have both lower + 2^split upper equal input or input+modulus. But when we would to strict comparison against the modulus, then the added overhead wouldn't be worth it to use as a separate method and should be keep using binary decomposition directly.

Additionally, there is additional bug in

gnark/std/math/bitslice/partition.go

Line 61 in 6ec0af7

rh.Check(upper, upperBound)

where we check

rh.Check(upper, upperBound)

instead of

rh.Check(upper, upperBound - int(split))