fix(configurator/bridge_mrp): address all bugs from code review

Two critical, one important, four polish fixes found by the
pr-review-toolkit code-reviewer.

C1 (CRITICAL) Start-at-node filter dropped later siblings
  fusion_plating_bridge_mrp/models/mrp_production.py:448
  The allowed_ids set was {descendants} ∪ {ancestors}, which wrongly
  excluded nodes that should run AFTER the start node — including
  later siblings of the start node and all operations in subsequent
  sub-processes. Rewrote the upward walk to ALSO include each
  ancestor's later-sequence siblings and their descendants. Smoke on
  ENP-ALUM-BASIC: full=9 WOs, partial from mid-tree 'De-Masking'=5
  WOs (previously was 1).

C2 (CRITICAL) Duplicate MO on re-confirm of pre-PR SOs
  fusion_plating_bridge_mrp/models/sale_order.py:96
  Legacy untagged MOs (created before this PR had line-linkage m2m)
  were not recognized by the untagged idempotency check, so
  re-confirming an already-processed SO would create one additional
  MO per untagged plating line. Fix: pre-scan for a single legacy
  untagged MO and adopt it by linking ALL untagged plating lines
  onto it. Those lines are then treated as covered and no per-line
  MOs are created on top. Smoke: S00066 before=1 MO, after
  re-run=1 MO.

I5 (IMPORTANT) push_to_defaults wrote to pre-bump revision
  fusion_plating_configurator/wizard/fp_direct_order_wizard.py:236
  When create_new_revision=True, _get_or_bump_revision() returned a
  new part record that got written to the SO line, but the
  post-confirm push_to_defaults loop re-read line.part_catalog_id
  (still the OLD rev) and wrote defaults there, defeating the whole
  point of "save as default". Fix: cache resolved parts in a dict
  keyed by wizard-line ID during the build loop, and use that cache
  in the push_to_defaults pass.

I3/I4/I6 (PERF) Computes lacked @api.depends and did per-record
  search_count / search queries
  fusion_plating_configurator/models/sale_order.py
  _compute_nav_counts, _compute_workorder_count, _compute_wo_completion
  now:
  - declare @api.depends
  - batch via read_group across the whole self recordset
  - rebuild {origin: counts} dicts and assign per record

M7 (MEDIUM) No savepoint around per-group MO creation
  fusion_plating_bridge_mrp/models/sale_order.py:_fp_auto_create_mo
  A mid-loop exception left group 1's MO persisted and aborted
  groups 2..N. Wrapped each group's create in SAVEPOINT/RELEASE/
  ROLLBACK TO SAVEPOINT so one bad group no longer corrupts state.

M8 (MEDIUM) Email 'opened' status false-positived on internal CC
  fusion_plating_configurator/models/sale_order.py:_compute_email_status
  Switched from 'any notification is_read' to 'customer partner has
  a read email notification on this SO'.

M9 (LOW) start_at_node_id domain silently empty when coating unset
  fusion_plating_configurator/wizard/fp_direct_order_line.py:94
  Changed `('parent_id', 'child_of', ...)` to
  `('id', 'child_of', ..., or 0)` and clarified the help text.

Regression smoke passed all checks on odoo-entech.

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

fusion_plating/fusion_plating_configurator/wizard/fp_direct_order_line.py

@@ -94,9 +94,11 @@ class FpDirectOrderLine(models.TransientModel):
     start_at_node_id = fields.Many2one(
         'fusion.plating.process.node',
         string='Start at Node',
-        domain="[('parent_id', 'child_of', coating_config_id and coating_config_id.recipe_id.id)]",
+        domain="[('id', 'child_of', coating_config_id and coating_config_id.recipe_id.id or 0)]",
         help='For re-work jobs: pick the recipe step where this job should '
-             'begin. Skips ancestor steps in the generated work order.',
+             'begin. Pick a coating first — nodes are scoped to its '
+             'recipe tree. Skips earlier steps in the generated WO but '
+             'keeps later siblings and sub-processes.',
     )
     is_one_off = fields.Boolean(
         string='One-off Part',

fusion_plating/fusion_plating_configurator/wizard/fp_direct_order_wizard.py

@@ -235,9 +235,13 @@ class FpDirectOrderWizard(models.TransientModel):
             'order_line': [],
         }
 
-        # 4. One SO line per wizard line
+        # 4. One SO line per wizard line. Cache resolved parts (post
+        # rev-bump) so the push-to-defaults pass writes to the right
+        # catalog entry.
+        resolved_parts = {}  # {wizard_line_id: resolved part record}
         for line in self.line_ids:
             part = line._get_or_bump_revision()
+            resolved_parts[line.id] = part
             header = '%s - %s Rev %s (x%d)' % (
                 line.coating_config_id.name,
                 part.name,
@@ -270,14 +274,14 @@ class FpDirectOrderWizard(models.TransientModel):
         so = self.env['sale.order'].create(so_vals)
         so.action_confirm()
 
-        # 6. Push-to-defaults (C4) — after the part has been resolved /
-        # revision-bumped, write coating + treatments back onto the part
-        # catalog entry so the next order inherits the same defaults.
+        # 6. Push-to-defaults (C4) — uses the resolved part cached
+        # during the build loop so rev-bumped lines write defaults to
+        # the NEW revision, not the pre-bump one.
         for line in self.line_ids:
-            if not line.push_to_defaults:
+            if not line.push_to_defaults or line.is_one_off:
                 continue
-            part = line.part_catalog_id
-            if not part or line.is_one_off:
+            part = resolved_parts.get(line.id) or line.part_catalog_id
+            if not part:
                 continue
             part.write({
                 'x_fc_default_coating_config_id': line.coating_config_id.id or False,