[Feature Request] Refactor integration-test architecture, taxonomy, and coverage strategy

Problem

The integration suite uses inconsistent setup and ownership patterns. Shared helpers hide file-level AVA hooks, the multi-codec harness runs hidden concurrent work, several large files cover unrelated behavior, and filenames carry an increasingly unclear mix of product area and execution requirements.

The repository also lacks a clear coverage strategy. Line coverage is useful for pure SDK code, but it does not show whether server capabilities, authentication modes, codec paths, or deterministic replay are exercised.

Proposed direction

• Define one preferred fixture pattern with explicit ownership of the environment, client, Worker, converter, task queue, setup, and teardown.
• Represent codec variants as independently named AVA cases with isolated routing and results.
• Organize tests by behavior or product domain and establish a practical suite-sizing guideline using case count and measured duration.
• Keep Cloud eligibility machine-checkable while the taxonomy evolves.
• Combine line and branch coverage with integration capability, runtime/authentication, scenario, and replay coverage.

Migration should be incremental. Mechanical file movement should remain separate from behavior changes, and coverage instrumentation should not be used for performance measurements.

Done when

• The preferred fixture API and migration guidance are documented.
• Parameterized cases have independent names, routing, failures, and cleanup.
• Test files follow an agreed taxonomy and sizing guideline.
• Coverage reports explain which behaviors each metric does and does not represent.
• Refactors preserve existing assertions and local/Cloud behavior.

Related work

• Parent tracker: #2381
• Cloud adaptation: #2326