Fix(chat): resolve None vendor_id silently dispatched to orchestrator

[Jean-Regis-M]

Closes #415

---

Problem

_call_start_workflow(vendor_id: int) has no runtime guard against None.
Python type hints are unenforced at runtime, so when the LLM emits null for
vendor_id, the value silently propagates into task_data and is dispatched
to the orchestrator no error, no signal, phantom-vendor failures downstream.

---

Root Cause

# finbot/agents/chat.py ~L198–201
if not self.background_tasks:
    return json.dumps({"error": "Workflow engine not available"})

# ← no guard here; vendor_id=None falls straight through
task_data: dict[str, Any] = {
    "vendor_id": vendor_id,   # None written into task payload
    ...
}

The background_tasks guard above demonstrates the existing validation
pattern it simply wasn't applied to vendor_id.

---

Fix

Two lines inserted immediately after the background_tasks guard, before any
state mutation or task dispatch:

if not self.background_tasks:
    return json.dumps({"error": "Workflow engine not available"})

if vendor_id is None:                                          # ← added
    return json.dumps({"error": "vendor_id is required"})     # ← added

from finbot.agents.runner import run_orchestrator_agent  # pylint: disable=import-outside-toplevel
child_workflow_id = f"wf_chat_{secrets.token_urlsafe(12)}"
task_data: dict[str, Any] = {

Why is None and not if not vendor_id?
0 is falsy but is a technically valid ID value. The guard targets only the
None case as specified in the issue.

---

Behavior Matrix

Input	Behavior
vendor_id=None	Guard triggers → {"error": "vendor_id is required"}
vendor_id=0	Passes guard (0 is not None); orchestrator handles semantics
vendor_id=-1	Passes guard; DB/orchestrator rejects invalid FK as before
vendor_id="abc"	Passes guard (not None); type error surfaces downstream out of scope
Valid int	Guard skipped entirely; zero behavior change

---

Impact

• No breaking changes, signature unchanged, return type (str) unchanged
• Backward compatible, all callers passing a valid int are unaffected
• No side effects return before any state mutation or dispatch
• Mirrors existing pattern structurally identical to the background_tasks guard

---

Testing

# Targeted regression
pytest tests/unit/agents/test_chat_assistant.py::TestBoundaryAndTypeValues::test_chat_boundary_011_vendor_id_none_flows_into_task_data -v

# Full suite
pytest tests/unit/agents/test_chat_assistant.py -v

Before fix vendor_id=None returns status='started', assertion fails:

result = await assistant._call_start_workflow("do something", vendor_id=None)
assert json.loads(result).get("error")  # FAILS

After fix deterministic early return:

result = await assistant._call_start_workflow("do something", vendor_id=None)
assert json.loads(result) == {"error": "vendor_id is required"}  # PASSES

All existing happy-path tests are unaffected; the guard is a new early-return
The branch that valid-vendor_id calls never reaches.

---

CI

Check	Status
Compiles	Pure Python, no syntax changes, no new imports
Lint	if vendor_id is None: matches surrounding style; no pylint violations
Tests	Only the boundary test changes assertion direction (fail → pass)
Dependencies	None changed
API contract	Method signature and return type unchanged

---

Tasks

• Identified root cause: missing runtime guard on vendor_id before task_data construction
• Inserted is None guard at earliest safe interception point (after background_tasks check, before any dispatch)
• Verified vendor_id=0 edge case is not blocked by the guard
• Confirmed zero changes to happy-path flow
• Confirmed method signature and return type are unchanged
• Unit test added covering vendor_id=None boundary case
• Full test suite passes with no regressions