fix: use VALIDATION_ERROR for invalid model ID in estimate_resources

Problem

estimate_resources() in kubeflow_mcp/trainer/api/planning.py (line 540-542) returns ErrorCode.SDK_ERROR for invalid model ID formats (e.g., "random gibberish"). It should return VALIDATION_ERROR instead.

AI agents interpret SDK_ERROR as a transient backend failure and retry the same bad input, instead of fixing it.

Other validations in the same function (num_workers, batch_size, quantization) correctly use VALIDATION_ERROR.

Proposed Fix

Modify the error handler at line 531-544 to return VALIDATION_ERROR for format errors and SDK_ERROR for actual API/network failures. Preserve the existing "did you mean" suggestion logic.

Acceptance Criteria

• Invalid model ID format returns VALIDATION_ERROR, not SDK_ERROR.
• API/network errors continue returning SDK_ERROR.
• Focused tests, linting, and full test suite pass.