Verify ReplicationHandler abort and async fetch lifecycle #259

Context

The one-shot leader/follower fetchindex workflow and follower read-back are verified on Solr 8.11.4, 9.11.0, and 10.1.0. The exported abortfetch command and any polling/status lifecycle remain outside that verified behavior; they must not be inferred from the one-shot fetch test.

Scope

• Determine the documented abortfetch semantics and whether the ReplicationHandler exposes a reliable observable completion/status signal for an in-flight fetch.
• Add isolated leader/follower integration coverage only where the lifecycle can be made deterministic and safe.
• Cover timeout, abort, and cleanup behavior without affecting shared integration cores.
• If a deterministic runtime test is not practical, document the limitation and keep the operation explicitly unverified.

Acceptance criteria

• Behavior is tested per pinned Solr version or explicitly marked deferred/unverified with rationale.
• Any async wait is bounded and cleanup cannot leave a shared core in a changed state.
• Replication docs, API coverage, compatibility docs, and api-inventory.json agree.
• Existing one-shot fetch verification remains separately stated.

Reference review (2026-11-04)

• Apache Solr 10.1 documents abortfetch as aborting an index copy, details as returning configuration/current status, and enablepoll/disablepoll as controlling follower polling: https://solr.apache.org/guide/solr/10_1/deployment-guide/user-managed-index-replication.html
• The public guide does not define a stable field-level status schema or a dedicated fetch-status polling command. A deterministic test must therefore establish a controlled in-progress transfer before invoking abortfetch, and must assert the resulting follower state rather than infer success from the command response.
• Keep the test isolated to disposable cores and bounded; if a deterministic in-progress state cannot be established reliably, document the limitation and leave live cancellation explicitly unverified.

Implementation progress (2026-11-04)

• PR test: verify ReplicationHandler abort lifecycle #262 adds an isolated leader/follower abort test using disposable cores and bounded polling.
• Initial CI run #650 exposed that the transfer completed before the test observed isReplicating=true on Solr 8.11.4. The test now configures the leader's maxWriteMBPerSec default to 0.05 to hold the transfer open deterministically before issuing abortfetch.
• The revised CI run is #654; keep this issue open until all pinned-version jobs pass and the docs/API inventory are reconciled.