Stop user-data errors leaking to Sentry from migrations

[premtsd-code]

Summary

Per-resource user errors thrown by Destinations/Appwrite.php were leaking to Sentry through the migrations worker. This PR fixes the routing at the library layer.

• Convert ~15 throw new Exception(...) user-error sites in create*() methods to setStatus + addError + return false. No exception bubbles for expected outcomes; the failure is recorded where detected.
• Per-resource wrap at import() becomes a pure safety net: record + rethrow. Library bugs and infra failures bubble through; user errors stay in the migration report.
• Row-flush block catches DuplicateException and StructureException from the database adapter and re-throws as Migration\Exception so the wrap classifies them as user-facing.
• Critical context: Migration\Exception is now the library's marker for "user-facing migration error". The Appwrite worker uses instanceof Migration\Exception to decide Sentry routing.

Motivation

Errors like "Document already exists", "Attribute already exists", "Cannot set default value for array column", "Invalid document structure: Missing required attribute X" were being thrown with code 0 or PDO codes (e.g. 23000), and the migration worker's code === 0 || >= 500 heuristic re-published every one of them to Sentry. ~80% of the noise on the appwrite-queue-migrations Sentry stream came from this path.

After this PR, expected user-data conditions never throw; only bugs/infra do.

Test plan

• Migration with a duplicate column → "Attribute already exists" appears in the migration report, no Sentry event
• Migration with a duplicate row → "Document already exists" in the report only
• Migration with malformed source data → "Invalid document structure" in the report only
• Library bug (synthetic TypeError) → recorded in report and reaches Sentry with full trace

[greptile-apps]

Greptile Summary

This PR converts ~15 throw new Exception(...) user-error sites in Destinations/Appwrite.php to setStatus + addError + return false, preventing expected migration failures (duplicate attributes, missing tables, invalid IDs, etc.) from ever throwing and leaking to Sentry. A new inner try/catch in the batch-flush block of createRecord() traps DuplicateException and StructureException from the database adapter with the same non-throwing pattern.

• User-facing errors in all create*() methods are now recorded in-place and return false, so they never reach import()'s catch block or the worker's Sentry routing logic.
• createField() and createIndex() infra-failure catch blocks now re-throw the raw \\Throwable instead of wrapping it as Migration\\Exception('Failed to create column/index'), preserving the original exception type and message in the migration report.
• import() itself was not modified; its catch (\\Throwable) still records errors without re-throwing, meaning all exceptions — user and infra alike — are ultimately absorbed there.

Confidence Score: 4/5

Safe to merge for the primary goal of silencing user-data errors; the claim that infrastructure failures now reach Sentry is not backed by the code.

The user-error suppression changes are consistent and correct throughout all create*() methods. The one gap is that import()'s catch block was not changed — it still swallows every exception without rethrowing — so the raw Throwable re-throws from createField() and createIndex() never escape to the worker's Sentry router. The PR achieves its noise-reduction goal, but the stated secondary goal of routing genuine infra failures to Sentry is not implemented.

src/Migration/Destinations/Appwrite.php — specifically the unchanged import() catch block (lines 402-417) and whether it should selectively rethrow non-user exceptions.

Important Files Changed

Filename	Overview
src/Migration/Destinations/Appwrite.php	~15 user-error throw sites converted to setStatus+addError+return false; createField/createIndex infra-failure catches now rethrow raw exceptions rather than wrapping as Migration\Exception; batch-flush block gains DuplicateException/StructureException catches; import() itself was not changed.

_{Reviews (5): Last reviewed commit: "Fix message: 'Column' -> 'Attribute' lim..." | Re-trigger Greptile}