Function normalize_schema_param 

pub fn normalize_schema_param(
    schema: Option<&Value>,
) -> Result<Option<Map<String, Value>>, McpError>

Normalize a raw MCP schema parameter value into the column-name → type map expected by apply_schema_override and parse_schema_override.

The MCP tool parameter is declared as Option<serde_json::Value> so the rmcp / schemars pipeline emits a permissive true JSON Schema. In practice some MCP clients forward this field as a JSON-encoded string rather than a raw JSON object — e.g. Windsurf/Cascade serializes {"postal_code": "TEXT"} as "\"{\\\"postal_code\\\": \\\"TEXT\\\"}\"". If we only accepted Value::Object (the old v.as_object().cloned() pattern) the override was silently dropped and ingest would fail with a confusing 22P02 invalid input syntax error from hyperd when a column that the user explicitly wanted TEXT stayed INT.

Accepted shapes:

• None / Some(Value::Null) — no override.
• Some(Value::Object(m)) — used directly.
• Some(Value::String(s)) — s is parsed as JSON; must decode to an object. A non-object payload (array, number, etc.) is rejected with SchemaMismatch so the caller gets a clear error rather than a silent no-op.

Any other shape is rejected with SchemaMismatch for the same reason.

Errors

• Returns ErrorCode::SchemaMismatch if a Value::String payload is non-empty but not valid JSON, or if it decodes to anything other than an object or null.
• Returns ErrorCode::SchemaMismatch for any other Value shape (boolean, number, array, etc.).