Bugfix

#1159 reported an issue where row deltas were empty in tables after their first update from schema. This was due to the notify calls for first and subsequent updates being on different code paths (with different invocations of notify), and the line that added the primary key to be tracked as a changed row was missing. This PR fixes the issue, and adds a litany of tests in Javascript and Python.

Additionally, this PR fixes a previously overlooked issue - though we enable delta calculation based on whether they are used in on_update or not, 0-sided contexts would always calculate step and row deltas regardless of whether they were enabled or not. This seemed to be a default - get_deltas_enabled would always return true, and set_deltas_enabled would be a no-op for zero sided contexts.

Because there is no public API to get a delta without going through on_update and enabling deltas, and considering that row deltas are used far more widely than cell deltas (and cell deltas aren't even implemented in Python), it seems to make sense for 0-sided contexts to have the same behavior as 1/2 sided contexts, which respect get_deltas_enabled.

A next step might be differentiating cell/row delta enablement inside the context, which means that a callback that requests a row delta would not also enable cell delta calculation. This would be an internal API, as the cell/row delta methods are, and might bring some performance improvements.

This PR also adds benchmarking for deltas across different contexts, which means we can use it to test out actual performance over different versions.

Testing

Tests added for JS and Python, for multiple context types, computed columns, indexed, etc.

Screenshots (if appropriate)

Checklist

• I have read the CONTRIBUTING.md and followed its Guidelines
• I have linted my code locally, following the project's code style
• I have tested my changes locally, and changes pass on the Azure Pipelines CI.