fix: stop excluding measurement fields whose names contain "id" from metric detection

[EAWRCEQA]

Problem

_detect_metric_field() in agentic_core/schema_inference.py:96 decides whether a numeric column should become an aggregatable metric in the auto-generated semantic layer. It excludes "ID-like" fields with a substring test:

if any(p in name_lower for p in ["_id", "id", "code", "编号", "year", "月", "日"]):
    return False

Because "id" is checked as a substring, it matches any field name that merely contains those two letters. Common measurement columns are silently dropped from metrics: width, humidity, grid_load, candidate_count, and similar. On data upload (POST /api/data/upload → infer_schema), these valid metrics never make it into the semantic layer, so the assistant can't aggregate them.

Note this is inconsistent with the sibling function _detect_id_field() (same file), which already matches ID patterns precisely via endswith / equality rather than substring.

Reproduction

Loading the module and calling infer_schema on genuine (non-ID, low-cardinality) float measurements:

metrics detected by infer_schema: ['总营收']
  humidity   type=float  is_metric=False is_id=False
  width      type=float  is_metric=False is_id=False
  revenue    type=float  is_metric=True  is_id=False

width and humidity have is_id=False — they are legitimate metrics — yet are excluded purely because their names contain id.

Fix

Match id / _id precisely (standalone name or _id suffix), mirroring _detect_id_field:

if name_lower == "id" or name_lower.endswith("_id"):
    return False
if any(p in name_lower for p in ["code", "编号", "year", "月", "日"]):
    return False

After the fix, the same input yields:

metrics detected by infer_schema: ['平均湿度', '总营收', '平均宽度']
  humidity  is_metric=True
  width     is_metric=True

and real ID columns are still excluded (id, vehicle_id, user_id → not metrics).

Verification

• python -m py_compile agentic_core/schema_inference.py passes.
• Ran the module directly (via importlib, so it doesn't pull the package's optional strands import) before and after the change; confirmed the excluded measurement fields are recovered and that id / *_id columns stay excluded.

The change is limited to the name-matching heuristic in _detect_metric_field; the code / 编号 / year / 月 / 日 matches are preserved unchanged.