[100M test]: Under minimal kv workload and just 2FTS query threads 50% of the queries are failing due to context deadline exceeded

Steps

1. 3KV and 6 FTS nodes cluster. Each of 16C, 64G. FTS RAM quota is set to 50G
2. Load 100M base64 encoded 1536 4096 dim vectors to kv.
3. Build FTS index on it.
4. Run 100 upserts per sec on KV.
5. Start 2 FTS query threads. It is observed that a lot of queries are timing out due to context deadline exceeded
6. System CPU/RAM used is pretty much in limits:
7. Memory used by FTS is also looking in good limit: