Preserve nanoeval max concurrency when resuming run sets

[sylvesterkaczmarek]

Summary

• preserve an existing run set's stored max_concurrency when the database is reopened
• initialize the default limit only when the metadata key does not yet exist
• keep the existing run_set_id metadata behavior unchanged

open_run_set_db() currently upserts max_concurrency with ON CONFLICT ... DO UPDATE, resetting it to 1_000_000 every time an existing run-set database is opened.

This conflicts directly with nanoeval.bin.concurrency, which stores a user-selected limit in the same metadata key. Reopening or resuming the run set therefore silently discards that value before workers read it, and even the concurrency command itself resets the previous value as part of opening the database.

The fix changes initialization to ON CONFLICT DO NOTHING, so new databases still receive the default while existing run sets retain their configured limit.

Regression coverage sets a custom limit, closes the run set, reopens the same run-set ID, and verifies the value is preserved.