feat(perf): Remove unused loads in mem2reg and last stores per functi…

…on (noir-lang#5925)

# Description

## Problem\*

Partially resolves noir-lang#4535.

This does not necessarily make SSA "aware of loops" but handles the
situation that we in the issue where we have multiple known stores and
loads that we are unable to resolve across the function due to loops.

## Summary\*

The per function state and the logic run over it has been expanded. We
store more information in `last_loads`, a new `load_results` map to
track unused loads, and a utility map `calls_reference_input` to avoid
deleting stores that are directly passed into an entry point with a
mutable reference.

To count the remaining stores, I reference count both the results of the
loads and the address of the loads. If the results of a load has a
counter equal to 0 I remove that load instruction. I then keep track of
each removed address with its own reference counter. If the removed
loads counter is equal to the counter of the address of the loads, we
know we have removed all loads to an address and thus can safely remove
the store to that address.

We also added a check that store we want to remove is not used as an
argument into a function call.

I have added a unit test `remove_unused_loads_and_stores` inside of
`mem2reg.rs`.

The `brillig_loop_size_regression` test:
```noir
unconstrained fn main() -> pub Field {
    let mut emulated_enum = EnumEmulation { a: Option::some(1), b: Option::none(), c: Option::none() };

    for _ in 0..1 {
        assert_eq(emulated_enum.a.unwrap(), 1);
    }

    emulated_enum.a = Option::some(2);
    emulated_enum.a.unwrap()
}
```
now compiles to the optimal SSA after a single run of mem2reg:
```
After Mem2Reg:
brillig fn main f0 {
  b0():
    v1 = allocate
    v2 = allocate
    v3 = allocate
    v4 = allocate
    v5 = allocate
    v6 = allocate
    jmp b1(u32 0)
  b1(v0: u32):
    v8 = eq v0, u32 0
    jmpif v8 then: b3, else: b2
  b3():
    v11 = add v0, u32 1
    jmp b1(v11)
  b2():
    return Field 2
}
```

## Additional Context

There is most likely other ways we can utilize this per function state.
We can continue to iterate upon it in follow-ups.

## Documentation\*

Check one:
- [X] No documentation needed.
- [ ] Documentation included in this PR.
- [ ] **[For Experimental Features]** Documentation to be submitted in a
separate PR.

# PR Checklist\*

- [X] I have tested the changes locally.
- [X] I have formatted the changes with [Prettier](https://prettier.io/)
and/or `cargo fmt` on default settings.

---------

Co-authored-by: Tom French <[email protected]>
Co-authored-by: jfecher <[email protected]>