test(streaming): raise streaming and broadcast coverage above 90%

## Goal

Raise Orleans streaming and broadcast-channel coverage above 90%, prioritizing serialization contracts, queue balancing, producer/consumer lifecycle, and failure paths.

Coverage report: https://github.com/dotnet/orleans/runs/98679059570

## Measured gaps

- `Orleans.Streaming` has roughly 840 uncovered physical source lines and is near 81% line coverage.
- Broadcast channels are near 65% line coverage.
- `AsyncStreamConverter.Read`: 0% coverage, complexity 24, CRAP 600.
- `EventSequenceTokenJsonConverter.Read`: 0% coverage, complexity 20, CRAP 420.
- `StreamId`/`InternalStreamId` JSON parsing and property-name paths have low branch coverage.
- Queue-balancer, subscription, producer/consumer lifecycle, cache, and error paths remain below target.

## Work

- [ ] Round-trip stream IDs, qualified IDs, async streams, and event sequence tokens in value and dictionary-key forms.
- [ ] Cover null namespaces, Unicode, event indexes, malformed JSON shapes, invalid discriminators/type mismatches, missing keyed providers, and default values.
- [ ] Exercise queue-balancer ownership changes and lease failure/recovery deterministically.
- [ ] Cover producer/consumer completion, error, unsubscribe, reconnect, and duplicate-subscription outcomes.
- [ ] Add focused broadcast-channel lifecycle and observer-failure coverage.
- [ ] Reuse fake providers and in-memory fixtures for contract tests; reserve external-provider behavior for provider-specific issues.
- [ ] Review reachable `NotImplementedException` paths such as producer completion/error callbacks and either implement or explicitly test the supported runtime contract.

## Acceptance criteria

- Streaming and broadcast packages reach at least 90% line coverage and materially improve branch coverage.
- Listed JSON converter methods reach at least 80% line and 70% branch coverage with CRAP below 30.
- Tests assert delivery, ordering, identity, retry, and lifecycle outcomes.
- Provider-specific service coverage remains in the cloud-provider issue to avoid duplication.

Start with the linked coverage report and the named AsyncStreamConverter.Read and EventSequenceTokenJsonConverter.Read paths, then inspect the existing fake providers and in-memory fixtures. Cover the listed serialization, lifecycle, queue-balancer, and failure cases; done means both streaming and broadcast packages reach 90% line coverage and the stated converter thresholds.

Written by the indexing model from the issue text.