Reduced shared parquet file contention in testing

[wilsonbb]

Change Description

Since moving to dask-espr, we've seen increased test failures. In #434 there have been recent failures in test_batch_by_band where a flux column can sometimes be read in as NaNs during the from_parquet call used in that test invocation. Interestingly this only seems to occur when the tests are run as a suite and not when test_batch_by_band is run in isolation.

We can achieve the same effect by having the test build an ensemble using duplicate files of the source and object files. This appears to fix the test when run on github actions, but we should still investigate the underlying issue with from_parquet and then revert this change.

• My PR includes a link to the issue that I am addressing

Code Quality

• My code builds (or compiles) cleanly without any errors or warnings
• My code contains relevant comments and necessary documentation

Bug Fix Checklist

• My fix includes a new test that breaks as a result of the bug (if possible)

[github-actions]

Before [8f33ee8]	After [1c6dc5a]	Ratio	Benchmark (Parameter)
33.0±0.8ms	32.4±0.1ms	0.98	benchmarks.time_prune_sync_workflow
33.2±0.6ms	32.0±0.7ms	0.96	benchmarks.time_batch

Click here to view all benchmarks.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 95.82%. Comparing base (40236a3) to head (5eb4149).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #440   +/-   ##
=======================================
  Coverage   95.82%   95.82%           
=======================================
  Files          25       25           
  Lines        1772     1772           
=======================================
  Hits         1698     1698           
  Misses         74       74           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[dougbrn]

Interesting that this works, makes me really want to have TAPE generate it's own test data via generator functions rather than use these test datasets...