Silence repeated rate limit and connection errors in OpenAI utils

[hemanth-asirvatham]

Motivation

• Reduce noisy repeated stack traces for common transient errors (rate limit, connection, and timeouts) while preserving first-occurrence visibility.
• Keep users informed via existing periodic status updates rather than spamming logs on every failure.

Description

• Added one-time logging flags (first_rate_limit_logged, first_connection_logged, first_timeout_logged) and helpers (_log_rate_limit_once, _log_connection_once, _log_embedding_rate_limit_once, _log_embedding_connection_once, _log_embedding_timeout_once) to track and log the first occurrence of rate limit/connection/timeout errors.
• Replaced noisy logger.warning/_emit_first_error calls in get_all_responses handlers for RateLimitError and APIConnectionError with the new one-time logging helpers so subsequent occurrences are logged at debug level.
• Applied the same first-occurrence silencing behavior to the embedding worker in get_all_embeddings for timeouts, rate limits, and connection errors.

Testing

• No automated tests were run for this change.

---

Codex Task

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}