…oss repeated format calls without freezing theme defaults, and route generic format aliases consistently through subplot and figure formatting paths. This fixes the dark-mode regressions on alternate axes, restores polar frame styling after the shared helper split, and adds regression coverage for persistence, clearing overrides, and theme transitions.

…n formatting all derive their generic alias handling from one shared schema instead of repeated manual key lists. This keeps the explicit override model intact while making format-key dispatch and persistence much easier to maintain as new axis styling options are added.

PR #725's refactor of axis-format routing dropped two real things:

* User-supplied lw= / ec= / edgecolor= (and axis-prefixed
  xlw=, xec=, …) no longer reach the spines, because
  `_merge_axis_style_state` only looks for the canonical `linewidth` /
  `color` names listed in `_AXIS_STYLE_FIELD_TEMPLATES`. The old
  `_update_background` path used to alias them via `_pop_props` and apply
  them to spines and ticks; that branch is gone now, so on this PR
  `format(ec='black', lw=1.4)` silently rendered default-thin default-color
  spines (visible in the test_uneven_span_subplots regression).

  Add the short aliases as additional sources for the `color` and
  `linewidth` schema fields so the existing merge logic picks them up.

* `_resolve_axis_format` has `tickwidth = get("tickwidth")` immediately
  followed by an unconditional reassignment of `tickwidth` from
  `prev`/`rc` fallbacks, throwing the `get("tickwidth")` value away. The
  result is the same in the test cases that exercise it today because
  `prev` carries the user-supplied value too, but the dead line is
  misleading and a foot-gun. Fold `get("tickwidth")` into the
  `_not_none` chain the same way `ticklen` / `tickwidthratio` already do.

1370 tests pass on the full suite.