Setting "resume_after" with a completed run results in "write_data_dictionary" step failing with multiprocessing

[Syonv]

Describe the bug
When running a partial model run using resume_after with multiprocessing enabled, and reusing the pipeline files of a completed model run, the write_data_dictionary step fails with the following error: RuntimeError: checkpoint 'write_data_dictionary' not in checkpoints. See traceback for more details: mp_summarize-activitysim.log. This error seems to stem from the fact that the checkpoints table (checkpoints.parquet in pipeline.parquetpipeline) is not overwritten from the resume_after model onwards and still holds references to checkpoints from beyond the mp_households step.

To Reproduce
Steps to reproduce the behavior:

1. Run the prototype_mtc model with multiprocessing
2. Set resume_after to some model
3. Re-run model in the same output directory as in step 1
4. See error

Expected behavior
The write_data_dictionary step should not load tables tagged with itself or beyond.

Additional context
This was tested with version 1.3.4 and the MWCOG model, as well as version 1.5.0 and prototype_mtc.