refactor: extract core/workflow/ subdirectory

[spinje]

Summary

Extract 6 workflow-related files from src/pflow/core/ into a new src/pflow/core/workflow/ subdirectory. Preparatory refactor for the planning module removal (Phase 1 places new files into core/workflow/).

Changes

File moves (via git mv, history preserved):

• workflow_manager.py → workflow/manager.py
• workflow_validator.py → workflow/validator.py
• workflow_save_service.py → workflow/save_service.py
• workflow_data_flow.py → workflow/data_flow.py
• workflow_status.py → workflow/status.py
• skill_service.py → workflow/skill_service.py

New files:

• core/workflow/__init__.py — re-exports all public symbols (22 symbols from 6 modules)
• core/workflow/CLAUDE.md — per-file details, internal dependencies, known issues, key lessons

Import updates:

• 36 src/ files — import paths updated
• 30 test files — import paths + 52 mock.patch() string targets updated
• 8 CLAUDE.md files — stale references updated

Cleanup:

• Removed dead re-exports from core/__init__.py (zero consumers for WorkflowValidator, CycleError, build_execution_order, validate_data_flow via that path)
• Reorganized core/CLAUDE.md — workflow-specific content moved to workflow/CLAUDE.md, parent keeps cross-cutting concerns

Also included:

• template_validation/CLAUDE.md — added Design Decisions section (two regex patterns, coalesce pre-splitting, type matrix, validation boundaries)
• .claude/commands/refactor.md — lessons learned: sed for bulk replacements, relative import auditing, intermediate checkpoints, __init__.py transitive loading, CLAUDE.md content placement

Explanation

core/ had 20+ files and was becoming cluttered. The 6 workflow files form a cohesive group (they cross-reference each other, share the same domain). Extracting them into a subdirectory reduces noise for agents working on unrelated core files (settings, metrics, shell integration) and gives the workflow domain its own CLAUDE.md for detailed documentation.

This is the pre-task for the planning module removal — Phase 1 of that work places new discovery functions into core/workflow/, so the subdirectory needs to exist first.

Testing

make test   → 4128 passed, 485 skipped (exact baseline match)
make check  → ruff, mypy, deptry all clean

Zero regressions. Stale reference sweep confirms zero old-path references in any .py or .md file.

[spinje]

Context: Why this refactor is needed

This PR is the pre-task for a larger effort: removing the gated planning/ module and repair system from pflow.

The bigger picture

The planning module (~7,300 production lines, ~23,000 test lines) and repair system (~1,050 production lines) were gated by Task 107 (markdown format migration). All code is preserved but unreachable — 516 skipped tests, entry-point guards everywhere. The architecture docs call the planner "legacy" and "being phased out."

Rather than maintaining this dormant code, we're removing it cleanly in 3 phases:

Phase	Scope
Pre-task (this PR)	Extract core/workflow/ subdirectory
Phase 1	Relocate active dependencies out of planning/ — move build_workflows_context to core/workflow/context.py, replace WorkflowDiscoveryNode with core/workflow/discovery.py, move context builders to registry/
Phase 2	Delete planning/, repair, gated code paths, ~516 skipped tests

Why extract core/workflow/ first?

Phase 1 places new files into core/workflow/ (workflow context builder, workflow discovery function). The subdirectory needs to exist before that work begins.

Doing the extraction separately (Decision #8 in the master plan) keeps the blast radius contained — 74 files touched, 52 mock.patch strings updated — and independently testable. Mixing this mechanical refactor with the planning removal logic changes would make both harder to review and debug.

Full planning docs: scratchpads/planning-removal/00-master-plan.md