[Perf]: Optimize InMemoryTaskStore::List for large task sets

[MisterVVP]

Benchmark signal

BM_TaskStore_List_ManyTasks is the largest hotspot by far, around 1.13 ms. It is much slower than point lookup benchmarks and also feeds into REST/JSON-RPC ListTasks paths.

Problem

The current list path likely does too much per request: scans ordered IDs, resolves map entries, filters, builds a temporary pointer vector, copies/project tasks, and paginates.

Task

Improve InMemoryTaskStore::List performance for 1k+ tasks.

Scope

• Avoid unnecessary temporary vectors where possible.
• Apply pagination earlier when filters allow it.
• Avoid copying full protobuf Task objects until the item is actually included in the response.
• Consider maintaining a compact ordered task index or cached sorted view.
• Benchmark filtered and unfiltered list paths separately.

Success criteria

• Reduce BM_TaskStore_List_ManyTasks by at least 40%.
• Target: from ~1.13 ms to ≤700 µs.
• Ensure BM_RestTransport_ListTasks and BM_JsonRpcTransport_ListTasks improve as follow-on effects.