…-packages#5502)

This PR refactors the GithubDependencyResolver to take a fetch
implementation instead of always using the global default

)

Resolves #4428

This is a followup to
AztecProtocol/aztec-packages#5341 which does the
initial ACIR generation work for multiple ACIR functions.

Execution is now done by moving `execute_circuit` to be part of a
stateful `ProgramExecutor` that builds a witness stack for every
completed `execute_circuit` call. An initial `execute_program` function
instantiates the `ProgramExecutor` and starts execution on our `main`
entry point circuit. When a `Call` opcode is reached we pause execution
and recursively call `execute_circuit`, which then returns the solved
witness for that call. We then resolve the outputs of that execution by
reading the return witnesses from the inner solved witness. We then push
the nested call solved witness onto the witness stack and continue
execution of our main ACVM instance. This is quite similar to the
process used by foreign calls with Brillig, except it is now done with
the main ACVM rather than the contained Brillig VM.

This witness stack and program (list of `Circuit` functions) then gets
passed to the backend. For now, I have only done an additive
`prove_and_verify_ultra_honk_program` to show the process working for
the basic non-inlined ACIR programs supplied here. I wanted to leave any
WASM exports or ACVM JS changes for a follow-up PR as they are quite a
bit of changes on their own.

---------

Co-authored-by: Tom French <15848336+TomAFrench@users.noreply.github.com>
Co-authored-by: Tom French <tom@tomfren.ch>
Co-authored-by: jfecher <jfecher11@gmail.com>

)

Resolves #4428

This is a followup to
AztecProtocol/aztec-packages#5341 which does the
initial ACIR generation work for multiple ACIR functions.

Execution is now done by moving `execute_circuit` to be part of a
stateful `ProgramExecutor` that builds a witness stack for every
completed `execute_circuit` call. An initial `execute_program` function
instantiates the `ProgramExecutor` and starts execution on our `main`
entry point circuit. When a `Call` opcode is reached we pause execution
and recursively call `execute_circuit`, which then returns the solved
witness for that call. We then resolve the outputs of that execution by
reading the return witnesses from the inner solved witness. We then push
the nested call solved witness onto the witness stack and continue
execution of our main ACVM instance. This is quite similar to the
process used by foreign calls with Brillig, except it is now done with
the main ACVM rather than the contained Brillig VM.

This witness stack and program (list of `Circuit` functions) then gets
passed to the backend. For now, I have only done an additive
`prove_and_verify_ultra_honk_program` to show the process working for
the basic non-inlined ACIR programs supplied here. I wanted to leave any
WASM exports or ACVM JS changes for a follow-up PR as they are quite a
bit of changes on their own.

---------

Co-authored-by: Tom French <15848336+TomAFrench@users.noreply.github.com>
Co-authored-by: Tom French <tom@tomfren.ch>
Co-authored-by: jfecher <jfecher11@gmail.com>