fix(postprocessing): correct boolean parsing in MemoryFileIterator CSV reader

[awanawana]

Summary

This PR fixes a real bug in MemoryFileIterator._format_csv() and adds the first test coverage for the skrl.utils.postprocessing module.

Bug Fix

File: skrl/utils/postprocessing.py

In Python, bool("False") evaluates to True because any non-empty string is truthy. This caused the CSV memory reader to silently corrupt boolean data — every "False" value in an exported memory CSV was read back as True.

# Before (buggy)
float(item) if item not in ["True", "False"] else bool(item)
# bool("False") == True  ← silent data corruption

# After (fixed)
float(item) if item not in ["True", "False"] else item == "True"
# item == "True" correctly maps "True" → True, "False" → False

This affects any user who exports a memory with boolean tensors (e.g. terminated, truncated) to CSV format and then reads it back with MemoryFileIterator.

Tests Added

File: tests/utils/test_postprocessing.py (new — 17 tests)

The skrl.utils.postprocessing module previously had zero test coverage. This PR adds comprehensive tests for MemoryFileIterator covering:

• NumPy (.npz) format: single file, multiple files (sorted order), multiple tensors
• PyTorch (.pt) format: single file, multiple files (sorted order)
• CSV format: float values, boolean True/False parsing (including regression tests for the bug), mixed booleans, invalid header, empty data rows
• Edge cases: empty glob, unsupported file extension, iterator exhaustion, for-loop usage, mixed-format directories

All 17 tests pass after the fix. The two boolean regression tests fail against the original code, confirming the bug.