feat(profiler): Reduce memory in Brillig execution flamegraph

[vezenovm]

Description

Problem*

No issue as these are local changes I had to enable executing larger flamegraphs

Summary*

Execution of the blob circuit when force compiled to Brillig produces 1.1 billion samples. The execution flamegraph process gets killed with an OOM error when running it on the mainframe which has 740G of RAM. The process would get killed when transforming the profiling samples returned by execute_program_with_profiling into the samples that the flamegraph generator expects.

To reduce memory I did a couple things:

1. We format our opcode directly. The AcirOrBrilligOpcode<F> enum was over 300 bytes. The opcode strings are mostly less than 20 characters.
2. Use std::mem::take when iterating over the profiling samples. This combined with step (1) got us to stop killing the profiling process. We still would use over 240 gigs.
3. Make a new Sample trait with a common interface. Then have different types of samples so that we are not storing unnecessary data. For example, in an execution sample the count will always be one. There is no need to store a count field in the sample. When you get to 1 billion samples this extra field starts to make a difference. This looks to have saved us over 20 gigs.

Additional Context

We can probably get rid of brillig_function_id as well by mapping the function ID to its samples.

Documentation*

Check one:

• No documentation needed.
• Documentation included in this PR.
• [For Experimental Features] Documentation to be submitted in a separate PR.

PR Checklist*

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[TomAFrench]

LGTM in theory, got some questions on some counter-intuitive bits plus suggestion