Disk buffer startup fails permanently when seek hits decode error (InvalidProtobufPayload)

Problem

On restart, disk buffer (v2) initialization can fail permanently when seek_to_next_record() hits a record that fails protobuf decode:

error occurred when building buffer
failed to seek to position where reader left off
failed to decoded record: InvalidProtobufPayload

Checksum, deserialization, and partial_write errors during initialization seek are treated as recoverable "bad reads", but ReaderError::Decode was not — so startup fails until the buffer directory is deleted manually.

Expected behavior

Decode failures during initialization seek should follow the same recovery path as other bad reads (skip and continue), not fail topology build permanently.

Related reports

• #18130, #19759, #25642 (same error family / seek-on-startup pattern)

Proposed fix

PR: https://github.com/vectordotdev/vector/pull/25691

Additional context

Tracked internally as Red Hat LOG-9468 (OpenShift Logging / AtLeastOnce disk buffer CrashLoop).

References

• Fixes #26231
• Related: #18130, #19759, #25642
• Internal tracker: https://redhat.atlassian.net/browse/LOG-9468

Start by reviewing PR #25691, then trace disk buffer v2 initialization through seek_to_next_record() and ReaderError::Decode. Compare decode-error handling with checksum, deserialization, and partial_write recovery; done means initialization skips the invalid record and continues instead of failing topology startup.