Retry Server long-poll capacity responses instead of stopping workers

[rmcdaniel]

Problem

A published Rust 2.0.0 worker registered successfully against a Cloud Dev 1-vCPU/1-GB cell, then exited when its activity poll received the Server's intentional 429 long_poll_capacity_exhausted response. The response was explicitly retryable and advertised retry_after_seconds: 1; PHP and Python workers completed equivalent journeys on the same cell.

A bounded Server admission response must not terminate an otherwise healthy Rust worker.

Acceptance criteria

• Workflow, activity, and query poll loops treat retryable 429 capacity responses as transient.
• Retry delay honors the response-provided retry hint where available while retaining bounded backoff.
• Non-retryable 4xx responses still fail promptly.
• Regression coverage proves the worker survives the capacity response and subsequently processes work.
• Publish the semver patch and rerun the managed Rust playground against the same Cloud Dev cell.

[rmcdaniel]

Implementation merged in #33 with public CI green. Keeping this issue open until 2.0.1 is published and the managed Rust playground completes against the same bounded Cloud Dev cell, as required by the acceptance criteria.

[rmcdaniel]

Acceptance is complete.

• durable-workflow 2.0.1 is published on crates.io from merged PR Handle typed poll capacity backpressure #33.
• Sample App PR Qualify Rust SDK 2.0.1 sample-app#80 pinned the published patch; its normal CI, polyglot, Compose, candidate-image, and public-boundary checks passed.
• The official ghcr.io/durable-workflow/sample-app-devcontainer:main image was rebuilt, qualified on amd64 and arm64, and published at digest sha256:cdb606de8c8b907eeef6605bc944284c09fdb64bba8e4828ddb1d77b44504b89.
• A clean origin/main checkout ran the Rust managed playground from that public image against the same 1-vCPU/1-GB Cloud Dev cell.
• Worker registration reported durable-workflow-rust/2.0.1; the workflow completed in 9.484 seconds with ActivityScheduled, ActivityStarted, ActivityCompleted, and WorkflowCompleted history.
• The result identified both workflow and activity runtimes as Rust.
• The temporary role-scoped client and worker credentials were revoked and removed immediately after the run.

This proves the retryable long-poll capacity response no longer terminates the published Rust worker, and all issue acceptance criteria are satisfied.