feat: add retry logic for empty LLM responses (CLI-49)

[szmania]

This PR implements retry logic for scenarios where the LLM returns an empty response.

Previously, an empty response would result in an error and immediate termination of the cecli execution. Now, users can enable retries for these cases via a new configuration option.

Changes:

• Added --retry-on-empty argument to cecli/args.py (default: False).
• Introduced EmptyResponseError in cecli/coders/base_coder.py to specifically handle empty LLM responses.
• Updated send_message in cecli/coders/base_coder.py to catch EmptyResponseError and perform retries if --retry-on-empty is enabled, using the existing exponential backoff strategy.
• Modified response handling in cecli/coders/base_coder.py to detect empty responses and raise EmptyResponseError.

This ensures better resilience when dealing with flaky LLM providers or local models (like Ollama) that might occasionally time out or return empty content.

Fixes CLI-49

[szmania]

drafted, need one more update

[szmania]

this is ready