combine overlapping sources in FIM, stable covariance

[gemmahuai]

In the original variance_update branch, the major failure mode is the inversion of Fisher information matrix (FIM) caused by perfectly overlapping sources. If there are any such sources in a single Tractor run, the matrix inversion would fail and we would lose information about other non-overlapping sources.

This PR fixes such a problem by combining overlapping sources into the a single entry (col/row) in the FIM so that the FIM is always invertible and we wouldn't lose info among other sources of interest.

Performance is summarized here.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 0% with 31 lines in your changes missing coverage. Please review.
✅ Project coverage is 45.73%. Comparing base (7c8e3c8) to head (1382520).
⚠️ Report is 1 commits behind head on variance_update.

Files with missing lines	Patch %	Lines
tractor/optimize.py	0.00%	31 Missing ⚠️
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@                 Coverage Diff                 @@
##           variance_update     #112      +/-   ##
===================================================
- Coverage            45.88%   45.73%   -0.15%     
===================================================
  Files                   27       27              
  Lines                 7103     7126      +23     
===================================================
  Hits                  3259     3259              
- Misses                3844     3867      +23     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[coveralls]

coverage: 45.735% (-0.1%) from 45.88%
when pulling 1382520 on covariance_fix
into 7c8e3c8 on variance_update.