feat: Log failed LLM calls to PromptLayer before re-raising

[adagradschool]

Summary

When .run() throws on an LLM call, the SDK now catches the exception, logs it to PromptLayer with status=ERROR, error_type, and error_message, then re-raises. Also adds these fields to log_request() for manual callers. Error categorization is duck-typed — no provider imports in runtime code.

Edge cases tested

• Tracking failure doesn't mask the original LLM exception
• Success path excludes error fields
• status_code=402 quota detection
• Substring matches ("quota", "timeout") gated to known provider modules only
• Real OpenAI/Anthropic exception classes

Known concerns

• Retry latency: Error-tracking in the except block uses track_request/atrack_request which have retry logic. If the PromptLayer API is down, this adds latency before the original exception propagates.
• throw_on_error interaction: When throw_on_error=True, a failure in the error-tracking call itself (e.g. PromptLayer API rejects the payload) would raise a PromptLayerAPIError inside the inner try/except, which we suppress to re-raise the original LLM error. This is intentional — the LLM error always takes priority — but means tracking failures are only visible at logger.debug level regardless of throw_on_error.

🤖 Generated with Claude Code

[hasaan21]

Update package version in promptlayer/init.py and pyproject