refactor: indirects and read/write slices (#7082)

This PR does a few things to clean up the indirect and mem handling in
the avm trace.

1) Adds support to resolve indirects across all operands in a similar
way to the simulator.
2) Introduces the `constrained_read` and `constrained_write` mem
abstractions that handle the (potential) indirect load step without
needing to manually write them.
3) Indirect handling in the `write_slice` and `read_slice` helpers.
4) Clean up gadgets read/write cycle - dropping unnecessary reads of
indirect destination offset thanks to above point
5) Simplified gadgets execution tests, removed checks other than return
result since they were arbitrary based on our read pattern at the time
and not reflective of what the tests should look like when they are
constrained.

Note the CMOV, MOV and JUMPI have been left for a later refactor as they
use specific load patterns in the mem trace, plus this PR was getting
large..