[BUG] auto_format_on_load crashes internally for valid RangeIndex time series #390

Auto-formatting failed: 'RangeIndex' object has no attribute 'freq'\n```\n\n## Reproduction\n```python\nfrom sktime_mcp.runtime.executor import Executor\n\nExecutor().load_data_source({\n    "type": "pandas",\n    "data": {"y": [1, 2, 3, 4]},\n})\n```\n\n## Current behavior\n- returns `success=True`\n- auto-formatting falls back after logging an internal failure\n- valid integer-indexed time series do not go through the normal formatting path\n\n## Expected behavior\n- integer / `RangeIndex` inputs should be treated as valid forecasting data\n- auto-formatting should skip datetime-only frequency inference for non-datetime indexes\n- metadata should preserve a sensible frequency label such as `Integer` instead of degrading or failing\n\n## Why this matters\nThe adapters and validation layer already allow integer-style indexes. The formatter should not assume datetime-only frequency metadata, otherwise a valid workflow partially fails behind the scenes and becomes harder for MCP clients to trust.