Resolve out-of-circuit max in PG recursive verifier

[lucasxia01]

Decsription update from Luke: We currently compute an out-of-circuit max of the log circuit sizes via get_max_log_circuit_size() in the PG recursive verifier. Aside from being incorrect, this results in needing to mock log circuit size in the dummy witness / VK construction context in certain unexpected cases, e.g. the hiding kernel. At the time of writing, it's not clear what the right solution is, but it seems likely that simply performing this max in-circuit is not sufficient.

[arielgabizon]

My view:max_log_circuit_size doesn't need to be present in the PG accumulator.
In the decider recurise verifier we can use CONST_PG_LOG_N instead of it.
This will make our use of PG more consistent with the paper, where the PG constraint system is globally fixed, it is not trimmed dynamically.

This related to another issue I wonder about.
I see currently in the decider recursive verifier we are padding up to a larger constant CONST_PROOF_SIZE_LOG_N which is 28 rather than the one above which is 20.
Seems potentially wasteful - if we are assuming the circuits being folded are of max size 20, why pad the accumulator up to size 28.