Remove leftover per-document bibliography blocks in panel_fixed_effects.ipynb and sensitivity_checks.md

[drbenvincent]

Summary

Remove the two remaining per-document :::{bibliography} blocks so the docs fully conform to the centralized-bibliography convention introduced by #834.

Background

PR #834 ("docs: fix Sphinx build warnings") consolidated all per-document bibliography blocks into a single global page at docs/source/references.rst, in order to eliminate bibtex.duplicate_citation and bibtex.duplicate_label warnings. Most notebooks/docs were migrated at the time, but two files still carry their own local bibliography directive:

• docs/source/notebooks/panel_fixed_effects.ipynb (added in Add PanelRegression experiment class for fixed effects estimation #628, after docs: fix Sphinx build warnings #834 merged)
• docs/source/notebooks/sensitivity_checks.md (lines 181–183)

Inline {cite:p} / {cite:t} citations in both files already resolve correctly against the global bibliography, so deleting the trailing :::{bibliography} directive is a pure cleanup with no behavioural impact.

Scope

• Remove the trailing ## References cell (containing only :::{bibliography} + :filter: docname in docnames + :::) from panel_fixed_effects.ipynb. Use notebook-aware tooling (nbformat, Jupyter, or the IDE notebook UI) so the JSON stays valid.
• Remove the equivalent ## References section + :::{bibliography} block at the bottom of sensitivity_checks.md.
• Re-execute panel_fixed_effects.ipynb end-to-end in the CausalPy env after the edit so the saved cell outputs match the new structure. Without a fresh run, the version checked in to git can render inconsistently on the remote ReadTheDocs build (stale outputs, mismatched cell numbering, etc.). sensitivity_checks.md is plain markdown and does not need execution.

Acceptance criteria

• grep -rn ":::{bibliography}" docs/source/notebooks docs/source/knowledgebase returns no matches.
• panel_fixed_effects.ipynb re-executed end-to-end in the CausalPy conda env; saved outputs are fresh.
• $CONDA_EXE run -n CausalPy make html (run from docs/) builds cleanly with no new warnings introduced and (ideally) no bibtex.duplicate_citation warnings emitted by these two files.
• Remote docs build (ReadTheDocs preview on the cleanup PR) renders both pages correctly, with citations resolving to the global references page.
• prek run --all-files passes (notably validate-notebooks for the .ipynb change).
• Inline {cite:p} / {cite:t} citations in both files still render correctly and link to the global references page.

Notes

• Independent of Add geo-experiment design workflow and sensitivity check plotting for Synthetic Control #819 and Refresh pipeline_workflow.ipynb and report_demo.ipynb outputs after #819 lands #883 — can be picked up at any time.
• A precedent for the notebook-side change is the recent commit on feature/sc-design-workflow that removed the same kind of cell from sc_pymc.ipynb (commit ba9c8d9).
• Suggested re-execution command: $CONDA_EXE run -n CausalPy jupyter nbconvert --to notebook --execute --inplace docs/source/notebooks/panel_fixed_effects.ipynb. If validate-notebooks fails afterward, follow the recovery steps in AGENTS.md.

[anevolbap]

Cleanup PR opened: #905.