fix: correct {step_N} placeholder regex in executor

[ShuxinLin]

Summary

Fixes #160.

_PLAN_PROMPT in planner.py is rendered with str.format(). The template contains {{step_N}} (double braces), which Python's .format() converts to {step_N} (single braces) in the prompt text the LLM receives. The LLM therefore generates args such as:

{"asset_id": "{step_2}"}

The old placeholder detection regex was:

_PLACEHOLDER_RE = re.compile(r"\{\{step_(\d+)\}\}")   # matches {{step_N}}

This never matched the single-brace {step_2} the LLM produced, so _has_placeholders() always returned False and the raw placeholder string was forwarded directly to the MCP tool — causing errors like:

{"error": "unknown asset_id {step_2} or no sensors found"}

Fix: change _PLACEHOLDER_RE to match single braces:

_PLACEHOLDER_RE = re.compile(r"\{step_(\d+)\}")        # matches {step_N}

All test fixtures updated from "{{step_N}}" to "{step_N}" to match actual LLM output format.

Test plan

• All 55 mcp/plan_execute/tests/ unit tests pass
• test_has_placeholders_true — "{step_1}" correctly detected
• test_executor_resolves_placeholder_via_llm — placeholder resolved via LLM call
• test_executor_no_placeholder_skips_llm — no-placeholder steps still skip LLM