opt_expr: fix sign extension for shifts

[georgerennie]

This fixes #5065. When optimizing a shift with constant shift amount, an optimization is used to limit the shift amount to the width of the input, but this limiting was done before potentially increasing the input's size by sign-extending it. This allowed the sign extended parts of the signal to be shifted into the output bits incorrectly. This PR defers the optimization until after the sign extension has happened to handle this and adds tests.

Could someone make sure that this doesn't break anything for $shift/$shiftx before merging please, I wasn't sure where they come from/what the difference is to the other shift cells.

[georgerennie]

Test failures seem to be that the testcase I added triggers UB in celledges

[georgerennie]

Have fixed that, seems this potential source of UB was known as there was b_width and b_width_capped but the former was still being used in some places. Switching to the latter everywhere fixes this (with the caveat in a comment above its description that if the shifted value is near 2**30 wide things will go wrong)

[KrystalDelusion]

Thanks @georgerennie ! Looks good to me.