Handle retry-after HTTP header (Groq API)

[Pipeliner]

Issue

Groq API has various limits, including tokens per minute: https://console.groq.com/docs/rate-limits.
When Aider encounter a 429 error, it does not use this header and implements its own delay scheme which does not work well.

Example:
Retrying in 1.0 seconds... 17:02:18 [24/1899]
litellm.RateLimitError: RateLimitError: GroqException - b'{"error":{"message":"Rate limit reached for
model gemma2-9b-it in organization ... service tier on_demand on
tokens per minute (TPM): Limit 15000, Used 11972, Requested 4351. Please try again in 5.289s. Visit
https://console.groq.com/docs/rate-limits for more
information.","type":"tokens","code":"rate_limit_exceeded"}}\n'
The API provider has rate limited you. Try again later or check your quotas.
Retrying in 2.0 seconds...
litellm.RateLimitError: RateLimitError: GroqException - b'{"error":{"message":"Rate limit reached for
model gemma2-9b-it in organization ... service tier on_demand on
tokens per minute (TPM): Limit 15000, Used 11390, Requested 4351. Please try again in 2.961s. Visit
https://console.groq.com/docs/rate-limits for more
information.","type":"tokens","code":"rate_limit_exceeded"}}\n'
The API provider has rate limited you. Try again later or check your quotas.
Retrying in 4.0 seconds...

Version and model info

v0.74.3.dev26+g185ea716
groq/gemma2-9b-it