refactor(operator-trend): collapse persistence/churn hotspot selectors onto a parametrized base (T3-2 phase 7a) (#91)

Pivot the fractal collapse to operator_resolution_trend.py (22K LOC).
Collapse the 4 reset-reentry persistence/churn hotspot selectors
(reset_reentry / rebuild_reentry / restore / rerestore) onto one
`_hotspots_base` + `_HotspotsTierSpec` + 4 specs. They were exact
structural clones differing only in the per-tier age / persistence / churn
keys, the `just-*` mode token, and the holding/sustained status set; the
churn-watch fallback set is identical across tiers and stays inline.

Proven byte-identical by an exhaustive branch differential (real vs
base+spec over a 9-class grouping / filter / sort / cap corpus across all
three modes per tier). This file has no composer-golden coverage, so the
differential plus the full suite (which exercises these via the public
resolution-trend API) are the gates.

Net -260 lines. Verification: pytest 2539 passed / 2 skipped; mypy clean;
ruff check clean.

Author: saagpatel