SK-3026: Align flowvault insert & detokenize with the FlowDB contract

Brings flowvault's insert and detokenize request/response classes in line with the FlowDB Java SDK Contract, reduces the module to bulk-only, and adds request-fidelity coverage.

52 files, +3,004 / −4,626 · 350 tests, 0 failures, 0 errors

Insert

Before | After -- | -- InsertRequest { table, upsert: List, upsertType, records } | { tableName, records: List, upsert: UpsertOptions } InsertRecord { table, data, upsert, upsertType } | new InsertRequestRecord { tableName, data, tokens, upsert } — adds BYOT tokens — | new UpsertOptions { String updateType, List uniqueColumns } InsertResponse { insertedFields, errors } | { records: List } — failures inline — | new InsertResponseRecord { tableName, skyflowId, fields, hashedData, httpCode, error } BulkInsertRequest (standalone) | extends InsertRequest, no fields of its own BulkInsertRecord | deleted → new BulkInsertRequestRecord extends InsertRequestRecord BulkInsertResponse { summary, success, errors } | { records: List, summary: BulkSummary } Success, Summary, Token | deleted — bulk-insert-only, replaced above

Tests

+55 request-fidelity tests asserting user input reaches the wire. assertSame on data/tokens/uniqueColumns proves pass-through rather than reserialization. Batching verified end-to-end with ArgumentCaptor over real dispatched batches: order preserved, vaultId and tokenGroupRedactions on every batch, response index matching original input position, and interceptor headers reaching each batch with a distinct RequestContext.

Non-ASCII values, spaced keys, and non-string data (Integer, Double, Boolean, nested Map/List) all verified to keep type and identity.

📄 docs/manual-test-scenarios.md — sections A–D cover flowvault (bulk insert, detokenize, tokenize/deleteTokens, headers & config).

Dependency

Depends on the common base classes becoming empty markers (BaseInsertRequest, BaseInsertResponse, BaseDetokenizeData) — flowvault extends them for future use. That change also touches v2 and breaks its japicmp gate; see the companion change for the baseline regeneration.

SK-3026: Align flowvault insert & detokenize with the FlowDB contract Brings flowvault's insert and detokenize request/response classes in line with the [FlowDB Java SDK Contract](https://skyflow.atlassian.net/wiki/spaces/SDK1/pages/3039166470/FlowDB+Java+SDK+Contract), reduces the module to bulk-only, and adds request-fidelity coverage.

52 files, +3,004 / −4,626 · 350 tests, 0 failures, 0 errors

Insert
Before After
InsertRequest { table, upsert: List, upsertType, records } { tableName, records: List, upsert: UpsertOptions }
InsertRecord { table, data, upsert, upsertType } new InsertRequestRecord { tableName, data, tokens, upsert } — adds BYOT tokens
— new UpsertOptions { String updateType, List uniqueColumns }
InsertResponse { insertedFields, errors } { records: List } — failures inline
— new InsertResponseRecord { tableName, skyflowId, fields, hashedData, httpCode, error }
BulkInsertRequest (standalone) extends InsertRequest, no fields of its own
BulkInsertRecord deleted → new BulkInsertRequestRecord extends InsertRequestRecord
BulkInsertResponse { summary, success, errors } { records: List, summary: BulkSummary }
Success, Summary, Token deleted — bulk-insert-only, replaced above
Detokenize
Before After
DetokenizeRequest { detokenizeData: List } { tokens: List, tokenGroupRedactions }
DetokenizeResponse { detokenizedFields, errors } { records: List }
DetokenizeRecordResponse deleted → new DetokenizeResponseRecord { token, value, tokenGroupName, metadata, httpCode, error }
BulkDetokenizeRequest (standalone) extends DetokenizeRequest
BulkDetokenizeResponse { summary, success, errors } { records: List, summary }
DetokenizeData, DetokenizeResponseObject, BulkTokenGroupRedactions deleted
The success/errors split is gone on both — failures are inline with per-record httpCode/error. Serialization moved from excludeFieldsWithoutExposeAnnotation() to serializeNulls() with transient internals, matching the doc's sample JSON.

Bulk-only
All unary methods removed from VaultController; the 8 bulk methods remain. Also removed: 16 orphaned classes (Get*, Query*, Tokenize*, DeleteTokens* request/response), 16 dead Utils/Validations methods, 36 unused imports, and implements IVaultController.

InsertResponse and DetokenizeResponse are retained as published API despite having no caller, since the contract specifies them.

⚠️ The doc still lists unary insert/detokenize/tokenize/deleteTokens as GA deliverables — confirm bulk-only is the intended scope.

Validation
Table name and upsert at exactly one level. tableName at request level or on every record, never both. upsert is optional but must match the table's level; it need not appear on every record.
updateType validated against FlowEnumUpdateType. Previously an unrecognized value ("MERGE", "update ") was silently dropped and the upsert ran with server defaults.
10,000-record cap enforced on all four bulk ops, using error constants that already existed unreferenced. Flips a prior test that asserted no cap.
UpsertType deleted — a hand-written duplicate of the generated FlowEnumUpdateType. Validator and mapper now read the same enum, so they can't drift.
Bug fixes
Fix Impact
NPE in getRecordsToRetry() / getTokensToRetry() A per-batch response with any 5xx record threw NPE
Blank tableName reaching the wire Validator treated " " as absent, mapper as present
Dead upsert wiring in insertBatchFutures Always-empty read; insertBatch always passed null
.exceptionally() side-channel Bulk insert/detokenize collected errors out-of-band and returned null
Tests
+55 request-fidelity tests asserting user input reaches the wire. assertSame on data/tokens/uniqueColumns proves pass-through rather than reserialization. Batching verified end-to-end with ArgumentCaptor over real dispatched batches: order preserved, vaultId and tokenGroupRedactions on every batch, response index matching original input position, and interceptor headers reaching each batch with a distinct RequestContext.

Non-ASCII values, spaced keys, and non-string data (Integer, Double, Boolean, nested Map/List) all verified to keep type and identity.

📄 docs/manual-test-scenarios.md — sections A–D cover flowvault (bulk insert, detokenize, tokenize/deleteTokens, headers & config).

Dependency
Depends on the common base classes becoming empty markers (BaseInsertRequest, BaseInsertResponse, BaseDetokenizeData) — flowvault extends them for future use. That change also touches v2 and breaks its japicmp gate; see the companion change for the baseline regeneration.