ZIR-194: Fix and test BigInt type inference

[tzerrell]

This fixes a number of problems in the type inference for BigInts in calculating the number of coefficients, their maximum positive & negative values, and their minimum bits. It also adds lit tests verifying these properties. For the lit tests, I manually computed the output types without looking at the algorithms to give some redundancy.

Also adds a verifier to the BigInt type enforcing:

• No overflow (i.e. maxPos + maxNeg < BabyBear)
• BigInts with a minimum value must be positive (i.e., at least one of maxNeg and minBits must be zero)

Makes the design decision that minBits should only be set if we know the number is positive, i.e. so that it is at least 2^minBits. Negative numbers don't count, even if we know they are large in absolute value.

I also stuck a rename into this PR since this forces a rebuild of all the BigInt files anyway -- the existing nondet_invmod vs inv vs ModularInvOp vs NondetInvModOp was confusing even me, and since we're in integers there's no real need to specify it's a modular inverse: here's no other sensible multiplicative inverse. The new names are nondet_inv (nondet) and inv (checked) (and corresponding variants, e.g. NondetInvOp and InvOp).

This could also be divided up as 3 PRs (one for BigInt type inference, one for the BigInt type verifier, and one for the renames of the inversion operations). I thought in this case it made more sense to unify them (in particular because that lets me make a single companion PR on the risc0 repo), but if we want to split them it would be straightforward as the commits are fairly clearly separated.

[pdg744]

Thanks for walking me through the details!
Might want to add a couple notes to help clarify:

• where to look to cross-reference the test logic & the algorithm that encodes that logic
• the subtleties around normalization we discussed

[jacobdweightman]

Awesome work. I love the new tests for the dialect, they're basically executable documentation! If you haven't already, it would probably be good to create an issue to track the Circom tests you turned off 😄

[tzerrell]

I'm going to wait for risc0/risc0#2413 to be approved before landing this, as otherwise we won't be able to bootstrap bigint into risc0 while this is landed and risc0/risc0#2413 is outstanding.

[tzerrell]

There's an unsigned commit snuck in the middle, so I'm rebasing to get everything signed.