Revisit brightfit init values and bounds

## Summary

Pulled real `tc_brightfit` ('bright' method) baseline-fit parameters directly
from AIND DocDB `quality_control` records for 10,401 fiber photometry
data assets (no raw traces needed — the fitted parameters are already stored
per ROI/channel). Used this to calibrate bleaching-rate assumptions in
`F0_OLS_vs_IRLS_vs_quantile-fit.ipynb` and to sanity-check the `tc_brightfit`
fitting code itself. Script: `pull_brightfit_params.py`.

## Method

- Filtered DocDB for `derived`-level, `fib`-modality, `behavior`-named assets.
- Extracted every `(asset, ROI, channel)` row's `b_inf/b_slow/b_fast/b_rapid/
b_bright/t_slow/t_fast/t_rapid/t_bright` from each session's stored
`Preprocessing using method 'bright'` QC evaluation.
- Derived `pct_decline = 1 − F0(session_end)/F0(0)` from the fitted curve
directly (`session_end` taken from the QC `Session length >15min` metric).
- Sanity-filtered before summarizing: `b_inf>0`, `60s