Add permission-aware follow-up flow with interactive resolution

When follow-up agent detects permission-blocked items ("ручное
применение", "permission denied", etc.), they become pending_actions
instead of auto-created tasks. User chooses per item:
  1. Rerun with --dangerously-skip-permissions
  2. Create manual task
  3. Skip

core/followup.py:
  _is_permission_blocked() — regex detection of 9 permission patterns
  generate_followups() returns {created, pending_actions}
  resolve_pending_action() — handles rerun/manual_task/skip

agents/runner.py:
  _run_claude(allow_write=True) adds --dangerously-skip-permissions
  run_agent/run_pipeline pass allow_write through

CLI: kin approve --followup — interactive 1/2/3 prompt per blocked item
API: POST /approve returns {needs_decision, pending_actions}
     POST /resolve resolves individual actions
Frontend: pending actions shown as cards with 3 buttons in approve modal

136 tests, all passing. Frontend builds clean.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

agents/runner.py

@@ -23,6 +23,7 @@ def run_agent(
     previous_output: str | None = None,
     brief_override: str | None = None,
     dry_run: bool = False,
+    allow_write: bool = False,
 ) -> dict:
     """Run a single Claude Code agent as a subprocess.
 
@@ -62,7 +63,8 @@ def run_agent(
 
     # Run claude subprocess
     start = time.monotonic()
-    result = _run_claude(prompt, model=model, working_dir=working_dir)
+    result = _run_claude(prompt, model=model, working_dir=working_dir,
+                         allow_write=allow_write)
     duration = int(time.monotonic() - start)
 
     # Parse output — ensure output_text is always a string for DB storage
@@ -106,6 +108,7 @@ def _run_claude(
     prompt: str,
     model: str = "sonnet",
     working_dir: str | None = None,
+    allow_write: bool = False,
 ) -> dict:
     """Execute claude CLI as subprocess. Returns dict with output, returncode, etc."""
     cmd = [
@@ -114,6 +117,8 @@ def _run_claude(
         "--output-format", "json",
         "--model", model,
     ]
+    if allow_write:
+        cmd.append("--dangerously-skip-permissions")
 
     try:
         proc = subprocess.run(
@@ -207,6 +212,7 @@ def run_pipeline(
     task_id: str,
     steps: list[dict],
     dry_run: bool = False,
+    allow_write: bool = False,
 ) -> dict:
     """Execute a multi-step pipeline of agents.
 
@@ -253,6 +259,7 @@ def run_pipeline(
             previous_output=previous_output,
             brief_override=brief,
             dry_run=dry_run,
+            allow_write=allow_write,
         )
         results.append(result)