Large start_pc leads to guest programs early exit

Since the tables are defined over a finite field, if the initial program counter is too large, the next program counter in the final row may wrap around to zero before the halt syscall is reached.

https://github.com/argumentcomputer/sphinx/blob/8a39b951e3ea520e295b693ad38bff6b43a2630c/prover/src/verify.rs#L93

As a result, a malicious prover can generate a proof that appears to show successful program termination, even though the halt syscall was never actually executed.

> This issue is disclosed with the permission of @johnchandlerburnham.

Start at prover/src/verify.rs:93 and trace how the verifier handles the initial program counter and the final row of the execution table. Reproduce the large-start_pc case, then verify that an execution cannot be accepted as successful termination when the halt syscall was not reached.

Written by the indexing model from the issue text.