plan executor: optimize_caps inherits the BGA-fanout options (skip its per-step reset)

Regression from the per-step reset (ef21a30): optimize_caps is the tail of the
BGA fanout, not a standalone op -- it clears decoupling caps from THE FANOUT'S
vias, at the fanout's clearance / via size (the CLI runs it as
place_fanout_clearance right after bga_fanout, with the same --clearance 0.09).
The plan's optimize_caps step carries no params of its own, so before ef21a30 it
correctly INHERITED the preceding fanout step's live control state. The per-step
reset wiped that -- optimize_caps then ran at the default clearance (0.25 instead
of 0.09) and stopped moving the caps: on rp2350 the CLI moves all 6 decoupling
caps (C18-C24, ~160-280um), while the GUI plan left them all at their original
positions.

Fix: skip the per-step reset for optimize_caps so it keeps the fanout step's
options (clearance/via/grid + the BGA cap-placement knobs). Leak-safe -- the
steps AFTER optimize_caps still reset, so nothing bleeds into the plane steps.

The board still graded 0 DRC before this (the #362 sync makes the router follow
the caps wherever they actually are), so this was a route-quality regression, not
a shorting one -- the caps just weren't getting their clearance optimization.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UKFKAEkBFgwTKh1E4oJkaQ

Author: drandyhaas

kicad_routing_plugin/claude_plan.py

@@ -831,7 +831,17 @@ def _next_step(self):
             # reset touches PARAMETERS only, not selections/log; apply_step_params
             # + apply_step_selection below then restore exactly THIS step's state.
             # (reset_params_to_defaults' own docstring says it is called here.)
-            if hasattr(self.dialog, 'reset_params_to_defaults'):
+            # EXCEPTION: optimize_caps is the tail of the BGA fanout, not a
+            # standalone op -- it clears decoupling caps from THE FANOUT'S vias,
+            # at the fanout's clearance/via-size (the CLI runs it as
+            # place_fanout_clearance right after bga_fanout with the same
+            # --clearance). It carries no params of its own, so it must INHERIT
+            # the preceding fanout step's options; resetting to defaults first
+            # runs it at the wrong clearance and it stops moving the caps. So
+            # skip the per-step reset for optimize_caps and let it keep the
+            # fanout step's live control state.
+            if (step["action"] != "optimize_caps"
+                    and hasattr(self.dialog, 'reset_params_to_defaults')):
                 try:
                     self.dialog.reset_params_to_defaults()
                 except Exception as _e: