Add exact writer epoch opens - #9

## Summary

- open every managed writer with the caller supplied nonzero authority
epoch
- rely on native SlateDB epoch fencing for lower-term, equal-term,
concurrent, and delayed stale opens
- add a request-local commit gate for bounded drain classification
without durable receipts or replay
- expose terminal retry classification across the Rust, TypeScript,
Python, and Go SDKs

## Stack

- SlateDB exact-epoch dependency:
HelixDB/slatedb#9
- pinned SlateDB commit: c64286ec8e5aaa724f12a8c54b58fe73efcff534
- consumed by Hyperscale:
HelixDB/helix-hyperscale#260

## Safety

- only a verified pre-execution rejection can return request ownership
- commit-start and drain-abort claims are mutually exclusive
- unknown write outcomes are terminal and never automatically retried
- authority exhaustion at u64::MAX remains intentionally unsupported

## Validation

Locked all-target/all-feature workspace compile passes on SlateDB 0.15.
Managed-writer lower/equal, concurrent, delayed stale, and in-flight
fencing tests pass. The locked migration parity dev suite passes
end-to-end. Fresh CI is rerunning after the final Python contract
formatting fix; full tests, Clippy, docs, SDK, and linked nine-node
validation remain in progress.

<!-- greptile_comment -->

<h3>Greptile Summary</h3>

The PR introduces exact-epoch managed-writer fencing, typed
reader-retirement cancellation, and terminal classification for unknown
commit outcomes. It also aligns the transport and SDK error contracts
and expands fencing, lifecycle, migration, and scale coverage.

- Managed writer opens pass a nonzero caller-supplied authority epoch to
SlateDB while embedded writer behavior remains separate.
- Reader retirement cooperatively cancels admitted reads with a
dedicated stable error code.
- HTTP reports unknown commit outcomes as unavailable with an explicit
non-retryable marker; gRPC reports them as unavailable with stable
metadata.
- Rust, TypeScript, Python, and Go SDKs expose explicit retry
classification without inferring retryability from status alone.
- CI and production contract coverage gain dedicated vector-migration
scale and writer-fencing checks.

<details><summary><h3>Important Files Changed</h3></summary>




| Filename | Overview |
|----------|----------|
| crates/db/src/lib.rs | Adds the managed-writer open mode, propagates
exact authority epochs to SlateDB, and exercises stale, equal,
concurrent, and in-flight fencing behavior. |
| crates/db/src/execution_control.rs | Adds monotonic reader-retirement
cancellation and integrates it with deadline-aware cooperative
execution. |
| crates/db/src/error.rs | Introduces typed retirement cancellation and
classifies fenced commit outcomes as terminal rather than transaction
conflicts. |
| crates/db/src/query_service.rs | Exposes predicates that let
transports distinguish reader retirement, deadlines, conflicts, and
unknown commit outcomes. |
| crates/server/src/http.rs | Maps unknown commit outcomes to HTTP 503
and emits an explicit `retryable: false` field. |
| crates/server/src/grpc.rs | Maps unknown commit outcomes to gRPC
Unavailable while retaining the stable error code in metadata. |
| sdks/rust/src/lib.rs | Preserves explicit retry metadata on remote
errors and retries only when the server supplies Boolean true. |
| sdks/typescript/src/index.ts | Adds optional retry metadata and
fail-closed retry classification to the TypeScript error contract. |
| sdks/python/src/helixdb/_client_common.py | Parses explicit retry
metadata in both current and legacy envelopes and exposes fail-closed
retry classification. |
| sdks/go/client.go | Adds optional retry metadata to remote errors and
requires explicit true for retryable classification. |

</details>


<details><summary><h3>Sequence Diagram</h3></summary>

```mermaid
sequenceDiagram
    participant C as Managed caller
    participant H as HelixDB
    participant S as SlateDB
    participant T as Transport
    participant SDK as Client SDK
    C->>H: Open writer with nonzero authority epoch
    H->>S: Open using exact writer epoch
    S-->>H: Writer opened or fenced
    C->>H: Execute write
    H->>S: Commit transaction
    alt commit succeeds
        S-->>H: Committed
        H-->>T: Success
    else fencing makes outcome unknown
        S-->>H: Fenced commit result
        H-->>T: Unknown outcome
        T-->>SDK: "503 / unavailable, retryable=false"
        SDK-->>C: Terminal error
    end
```
</details>

<sub>Reviews (1): Last reviewed commit: ["Accept disjoint membership
storage in
re..."](85e6a97)
| [Re-trigger
Greptile](https://app.greptile.com/api/retrigger?id=58590933)</sub>

<details><summary><h4>Context used (4)</h4></summary>

- Knowledge Base — [Database
engine](https://app.greptile.com/helixdb/-/custom-context/knowledge-base/helixdb/helix-db/-/docs/database-engine.md)
- Knowledge Base — [Interpreter
execution](https://app.greptile.com/helixdb/-/custom-context/knowledge-base/helixdb/helix-db/-/docs/interpreter-execution.md)
- Knowledge Base — [Client SDKs and
bindings](https://app.greptile.com/helixdb/-/custom-context/knowledge-base/helixdb/helix-db/-/docs/client-sdks-and-bindings.md)
- Knowledge Base — [Server and
transports](https://app.greptile.com/helixdb/-/custom-context/knowledge-base/helixdb/helix-db/-/docs/server-and-transports.md)
</details>


<!-- /greptile_comment -->