fix(checkpoints): preserve cumulative transforms across repeated saves

[nodesagar]

Summary

This PR fixes a checkpoint/history regression where a later save could silently discard transforms that were already included in an earlier checkpoint.

Before this patch, POST /projects/{project_id}/save rebuilt project state from the original CSV plus only the unapplied change-log entries. In a flow like renameCol -> save -> addCol -> save, the second save dropped the earlier rename because only the pending addCol log was replayed.

This patch makes checkpoint creation preserve the current working copy as the source of truth and only use save to mark pending logs as applied.

Closes #224.

What changed

• updated save_project() to read the current working copy instead of rebuilding from only the unapplied log tail
• added a focused regression test covering the two-save case where an earlier rename must survive a later save
• replaced the assert in the save path with an explicit 500 error if the working copy path is misconfigured

Validation

• ruff check .
• ruff format --check .
• pytest tests/test_save_revert.py -q
• pytest -q -k 'not test_add_column_without_name_returns_422'
• npm run lint
• npm run test -- --run
• manual two-save repro confirmed the final saved columns remain ['name', 'years', 'city', 'country']
• manual revert repro confirmed the latest checkpoint still restores ['name', 'years', 'city', 'country']

Note on baseline

main currently has one unrelated backend test failure in tests/test_new_features.py::TestAddDeleteColumnEndpoint::test_add_column_without_name_returns_422, which belongs to the separate addCol validation bug fixed in PR #221. Excluding that known baseline failure, the remaining backend suite passed locally on this branch.