[Architecture] Define the worldgen provider concurrency contract before enabling parallel provider execution

[ogyrec-o]

Problem / Motivation

The current v1 behavior is intentionally conservative and correct:
each hosted worldgen provider session executes through a serial execution domain.

That is the right first step for off-main-thread worldgen, but it is not yet a
contract for safe parallel provider execution. Before Freven runs multiple
columns concurrently for the same provider/session, it needs explicit answers
for:

• reentrancy expectations
• mutable session coupling
• determinism across worker counts / CPU counts
• provider fault boundaries
• cancellation / stale completion semantics under wider concurrency

Send + Sync alone is not enough to define gameplay/runtime correctness.

Target direction

Design and document an explicit concurrency contract for worldgen providers.

Possible acceptable directions:

1. stateless provider / factory-per-job model
2. explicit parallel-safe capability with documented guarantees

The result must say clearly:

• what concurrency modes exist
• what determinism guarantees exist
• what the host is allowed to schedule in parallel
• how faults / stale completions / shutdown interact with the model

Non-goals

• do not silently enable parallel execution just because it benchmarks faster
• do not treat the current serial lane as a temporary hack to bypass
  architecture work
• do not widen guest threading generally

Done when

• provider concurrency modes are explicit
• determinism expectations are documented and testable
• future parallel worldgen work has a real contract to build on