feat: Modify inter-shot random_seed behaviour, provide legacy behaviour through seed_mode

[jake-arkinstall]

Adjusts the random_seed behaviour to avoid overlap between batches of run_shots using nearby seeds.

The existing behaviour was such that if one ran run_shots of N shots with a random_seed of M, components are seeded with M on the first shot, M+1 on the second, up to M+N-1. This acts as a reproducibility aid, but causes confusion when expected entirely different batches of shots.

The new default behaviour is to randomise the seed per-shot, by passing random_seed to a Pcg64Mcg pRNG and yielding shot RNGs by advancing this RNG to the required shot. As such, the chance of any overlap between two subsequent run_shots runs is very small, even if the seeds are close.

The existing behaviour still serves some purposes, so is available via a new seed_mode parameter that can be provided to run_shots and run. Pass "legacy" for the existing mode, "default" for the new randomised mode. We demonstrate this by persisting some original snapshots with this mode in this PR, but may revise them to use the default mode later on.

The legacy mode will be supported throughout this major version. It will eventually be dropped entirely, and our only mode will be to randomise seeds between shots. To repeat prior experiments, shot_increment and shot_offset can be used as demonstrated in determinism tests.

[jake-arkinstall]

Some guidance for reviewers.

• The test/resources directory is just generated QIS. Some files were added when regenerating some guppy code for different seed_mode tests, but are widely irrelevant for the purposes of review.

• Notice that traces and result snapshots are unchanged compared to the original snapshots when giving them seed_mode="legacy". Note that some have simply moved when pytest.mark.parametrizing them over seed mode.

• I am not convinced that legacy and default are the right names, as they aren't descriptive, but as the legacy mode is to be dropped later on it seems reasonable enough.

• Note that a new test has been added comparing the seed modes. It is arguably the most important aspect of this PR to review.

[Copilot]

Copilot wasn't able to review this pull request because it exceeds the maximum number of lines (20,000). Try reducing the number of changed lines and requesting a review from Copilot again.

[cqc-alec]

Worth considering using an enum for this instead of a string?

[jake-arkinstall]

It could be. Though it will necessitate an extra import e.g.:

from selene_sim import SeedMode

run_shots(..., seed_mode=SeedMode.legacy)

[cqc-alec]

Yes, perhaps not worth it.