[FEATURE] Support initial rate boost in adaptive rate limiter

[dai-chen]

Is your feature request related to a problem?

With adaptive rate limiter enabled, the first bulk request per core can be delayed if a low minRate is configured. For example, with batch_size=3000 and minRate=1000, each first request must wait 3000 ÷ 1000 = 3 seconds. This leads to noticeable delays at the start of the ingestion.

What solution would you like?

One option is to introduce a configurable initialRate (similar to TCP Slow Start) for each rate limiter instance — this allows fast permit acquisition for the first few bulk requests, even when minRate is conservatively low. Once throttling is detected (high latency or retries), the limiter switches to normal congestion avoidance using AIMD algorithm.

What alternatives have you considered?

• Relying on a higher minRate, but it becomes too aggressive post-throttling.
• Reducing batchSize, but may increases overhead and reduce throughput.

Do you have any additional context?

N/A