Add benchmark on collection load time

[tellet-q]

• Introduce BENCHMARK_STRATEGY env variable to control the experiments, possible values: default | tenants | collection-reload
• Add a new server config qdrant-continuous-benchmarks-snapshot with QDRANT_NUM_CPUS: 4 to ensure no optimizations are triggered after snapshot recovery
• Add a new column collection_load_time_ms into database. Fetch the value from /telemetry response.

Prepare snapshots for cases, upload them to GCP&authuser=0&project=key-surf-337615):

• 500k points with all possible payloads, 768x, all default values
• 500k points with all possible payloads, 768x, on_disk:true (all possible on_disk places)
• 500k points with all possible payloads, sparse vectors, all default values
• 500k points with all possible payloads, sparse vectors, on_disk:true (all possible on_disk places)

[tellet-q]

Example run: https://github.com/qdrant/vector-db-benchmark/actions/runs/11035515022

[tellet-q]

@KShivendu @generall I wanted to run the runLoadTimeBenchmark job in parallel, almost everything is ready in code for that, but it not possible since the 2nd server has ARM architecture and we don't have Docker images for ARM.

[tellet-q]

I updated the code to run the benchmark in parallel with tenants benchmark. The benchmark runs in a separate server machine and only the client machine is shared.

https://github.com/qdrant/vector-db-benchmark/actions/runs/11051637232