Feat/1700 1701 1702 1703 test coverage

[a-malik-gh]

Four test-coverage fixes across credits, payments, wallets, and auth:

#1700 — atomic revenue-split update validation
RevenueSplitRecipientDto validates each recipient's basis points at the
DTO layer, but nothing proved RevenueSplitService.replaceRecipients
(backing PUT .../recipients) rejects a set that individually passes
per-field validation but doesn't sum to 10000. Added a test with two such
sets (9999 and 10001 total) and confirmed neither rejected update deletes
the existing recipients first — the validation already ran before the
transaction starts; this just proves it.

#1701 — reconciliation overlap protection + full-batch test
ReconciliationService had no bound on how long a full
PAYMENT_RECONCILE_MAX_BATCH pass could take against a slow provider —
500 sequential awaits up to PAYMENT_VERIFY_TIMEOUT_MS each can exceed the
5-minute cron interval. Added a process-local in-progress flag so an
overlapping @Cron tick becomes a no-op instead of double-processing (and
a test proving it), plus a test running a full 500-payment batch to prove
the PAYMENT_RECONCILE_MAX_BATCH cap doesn't silently drop candidates.
True wall-clock timing against a real provider is a staging soak-test
concern, not a synchronous unit test — the overlap guard is the structural
fix for the case where a pass does run long.

#1702 — concurrent custodial wallet provisioning
WalletsService.provisionCustodialWallet's doc comment already claims the
same first-write-wins (user_id) constraint pattern the ledger's overdraft
race gets a dedicated test for, but nothing exercised real concurrency.
Added a test firing 5 genuinely concurrent calls through a small in-memory
WalletAccount table (serializing transactions and enforcing the unique
index, mirroring credits/testing/in-memory-ledger.ts's own approach) —
exactly one wallet is created, every caller converges on it, and
KeyCustodyService.provisionKeypair is only ever called once.

#1703 — RolesGuard none-of-several case
Only the single-role case and "holds one of several" were tested. Added
the missing "holds none of several accepted roles" case. Since UserRole
currently has only two values (USER, ADMIN), a requiredRoles list
naming both is satisfied by any real user — so this is only reachable via
a role value outside the enum (stale JWT, bad seed data), which is exactly
the case the guard must not silently let through. "Holds more than one
role at once" isn't representable today (RequestUser.role is a single
UserRole, not UserRole[]) — documented explicitly in a comment per the
issue's own fallback instruction, rather than left ambiguous.

One additional out-of-scope commit: credits.service.ts and
credits.service.spec.ts on main had a stray fragment of the spec's
build() helper spliced into the middle of CreditsService itself
(between getSystemAccount and getPayableAccount), with the spec file
left holding an orphaned, syntactically-invalid second copy of the same
helper. This broke TypeScript compilation for the whole package — nothing
in credits/, including the #1700 test above, could type-check without
it. Removed the stray fragments; no behavioral change.

Tests / checks performed

• src/credits/revenue-split.service.spec.ts — all tests pass
• src/payments/reconciliation.service.spec.ts — all tests pass (overlap-guard tests confirmed via log output: "Skipping reconciliation pass: the previous pass is still running")
• src/auth/guards/roles.guard.spec.ts — all tests pass
• npx tsc --noEmit — zero errors in any file this PR touches (confirmed after the credits.service.ts fix; a handful of pre-existing, unrelated errors remain elsewhere in the repo — missing modules for credits-admin.service/payments-admin.service, mismatched controller/DTO signatures in unrelated *.controller.spec.ts files, an Express cookie overload mismatch in auth.controller.ts — none of them touch anything this PR changes)
• src/wallets/wallets.service.spec.ts — could not get a clean local run. This sandbox's npm registry connectivity was severely degraded for this session (confirmed via ~12 install attempts across npm install, npm ci, varied retry/timeout/concurrency settings, and detached/nohup runs — each failed a different way: ECONNRESET, ENOTEMPTY/EPERM on a nested rxjs cleanup during dedup, npm's own "Exit handler never called" bug, or a genuine multi-minute stall with zero progress). The one clean install I did briefly get before a later reinstall attempt wiped it showed all 3 other suites above passing 45/45, and this file's only compile blocker was a corrupted @stellar/stellar-sdk/@stellar/stellar-base install (an unrelated third-party package, not this PR's code) — reinstalling just that package fixed the compile error and the file ran, before the environment degraded further. I'm confident in the test's correctness (same in-memory-transaction-table pattern as the ledger's own passing overdraft-race test) but would appreciate CI confirming this suite specifically.

Closes

Closes #1700
Closes #1701
Closes #1702
Closes #1703

[vercel]

@A6dulmalik is attempting to deploy a commit to the naijabuz's projects Team on Vercel.

A member of the Team first needs to authorize it.

[drips-wave]

@a-malik-gh Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits

[yusuftomilola]

No merge conflicts with main. Good test coverage additions across roles guard, revenue-split, reconciliation, and wallets services - approving.