Remove non-functional joblib parallelization from TrendFit (#428)

* docs(plans): add private-dev + public-release repo plan

Documents the two-repo architecture for private development with
public releases via rsync-based export script.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* refactor(fitfunctions): remove joblib parallelization from TrendFit

Joblib's loky backend deadlocks on macOS with Python 3.12's spawn
start method. The multiprocessing backend fails due to unpicklable
closures, and the threading backend provides no CPU-bound speedup.

Remove the non-functional parallel code path, related tests,
benchmarks, and the performance optional dependency. All Phase 4
features in core.py (residuals use_all, in-place mask ops,
xoutside/youtside) are preserved — they have zero joblib dependency.

Generated with Claude Code

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: blalterman