Lookup serialization

[Nashtare]

This PR updates the way lookup tables are being serialized, and more generally gives access to CommonCircuitData to gates and generators during serialization/deserialization.

This allows:

• to serialize only once a lookup_table in a given circuit
• to remove the need for the special DummyCircuitGenerator to be placed first in the GeneratorSerializer object (could have easily been source of errors for external users defining their own gate/generators and needing to implement a custom serializer.

For this, it updates the underlying F type of WitnessGeneratorRef from F: Field to F: RichField + Extendable<D>. I am not sure as to why it was set only to Field in the first place, as the restriction is natural here.

[Nashtare]

The CI errors are not related to these changes. See #1113 .

[npwardberkeley]

Same with this

[Nashtare]

common_data is actually being used in quite a lot of places (and consistently across this file). Do you want me to address this particular case still?

[npwardberkeley]

Hm maybe it should just be called common_data everywhere (instead of cd)?

[Nashtare]

Ok! Addressed in additional commit.

[Nashtare]

@npwardberkeley I've added another commit on top of what you already reviewed, addressing a point mentioned in #1114 about circuit building which can take quite some time when we are loaded with lookups.
The commit adds a hash (keccak as already present) of the lookup table when defining a new LookupGate / LookupTableGate, and uses a custom implementation of Gate::id() for those two gate types, to ignore the entire table.

For 8192 LookupGates, I am getting a circuit building time of 2.45 sec vs 4.5 sec for the current implementation. Let me know what you think! 🙂

[npwardberkeley]

Looks good!