fix(demos): drop CSV cache from peak_streamflow_trends; run analysis live

The notebook had its ``peak_trend_analysis`` invocation commented out
with a "few minutes to run" warning, and the last cell loaded a
pre-computed ``datasets/peak_discharge_trends.csv`` instead. The
relative path broke whenever the notebook was executed from any cwd
other than ``demos/`` (e.g. an nbconvert run against a staged copy).

With the async chunker, the analysis runs end-to-end in roughly two
seconds for a single state's worth of stream gages, so the cached CSV
no longer earns its keep:

- Un-commented the live ``peak_trend_analysis`` call in cell 11 and
  reduced its scope from three states to Rhode Island. Rhode Island has
  ~350 stream gages — enough to actually exercise the chunker, but
  small enough that the doc build (which executes notebooks live under
  nbsphinx, unauthenticated) stays well within rate limits.
- Deleted the ``pd.read_csv("datasets/peak_discharge_trends.csv")`` cell
  (no longer needed; ``final_df`` is produced live).
- Deleted ``demos/datasets/peak_discharge_trends.csv`` (535 KB tracked
  cache) and the now-empty ``demos/datasets/`` directory.

The notebook now has no relative-path data dependency — it executes
cleanly from any cwd, including under nbsphinx in CI.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: thodson-usgs