Address sample_sequence_length greater than min_length_time_axis

[mickvangelderen]

Based on #43, please review that first and if it wasn't merged yet, please review this PR per commit.

The goal of this PR is to make the tests not generate flashbax warnings and to prevent code from being added that would trigger a flashbax warning.

I am not sure that the suggested changes in the last commit are right.

I am unsure why min_length_time_axis=min_length // add_batch_size + 1 is used in the trajectory buffer source code, particularly the + 1. I am also unsure why the min_length fixture is defined as:

@pytest.fixture
def min_length(sample_batch_size: int) -> int:
    return int(sample_batch_size + 1)

I could understand if it was defined in terms of add_batch_size, but it is not.

[SimonDuToit]

I think the + 1 in min_length_time_axis=min_length // add_batch_size + 1 is just to be conservative. This matches how max_length_time_axis is calculated.
The min_length fixture is defined to be viable, since it has to be at least as big as sample_size.

[SimonDuToit]

I'm not sure why its necessary to replace min_length with add_batch_size. Otherwise it makes sense, please just see my comment on the merge conflict.

[SimonDuToit]

Just two tests still failing. You forgot to change a line on one, and another one you changed unnecessarily.

[SimonDuToit]

Suggested change

	fake_batch = get_fake_batch(fake_transition, int(min_length + 10))
	fake_batch = get_fake_batch(fake_transition, int(min_length))

[SimonDuToit]

Please undo the changes to this test, test_add_sequences_and_batches. The problem you found only comes up when add_batch_size is manually set to something bigger than min_length, but here it just uses the argument which is fine.