fix(cli): stop tests from leaking jest's own exit code

With coverage finally over the line, the Unit Tests job stayed red — 1250
tests passing, branches 75.54% against a 75% threshold, and jest still
exiting 1 with no failure printed anywhere.

The difference was `--runInBand`, which is how CI runs and how nobody runs
locally. Without workers every suite shares jest's process, so a command
that sets `process.exitCode = 1` to carry its verdict (ADR-0073) leaves it
set for the rest of the run. In worker mode the same test is harmless,
which is exactly why this only ever appeared in CI.

Worth stating plainly: the leak in `scaffold.command.spec.ts` PREDATES this
session's coverage work — it reproduces on HEAD~1. So the Unit Tests job
could not have gone green even at 100% coverage; the coverage gate was
taking the blame for two failures while hiding one of them. The `update`
spec picked up the same leak from the tests added yesterday.

Fixed once, in `setupFilesAfterEnv`, rather than in the two specs that
happen to leak today. A unit test setting the runner's exit code is always
an artifact and never a signal — jest reports real failures on its own —
so restoring it after every test makes the whole class impossible instead
of waiting for the next command to reintroduce it.

Refs GT-562

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: beyondnetPeru

src/sdk/cli/jest.config.js

@@ -44,7 +44,7 @@ module.exports = {
     '^conf$': '<rootDir>/src/test/__mocks__/conf.ts',
     '^chokidar$': '<rootDir>/src/test/__mocks__/chokidar.ts',
   },
-  setupFilesAfterEnv: [],
+  setupFilesAfterEnv: ['<rootDir>/src/test/reset-exit-code.setup.ts'],
   testTimeout: 10000,
   verbose: true,
   // OPP-003: Suppress console noise during tests for cleaner output

src/sdk/cli/src/test/reset-exit-code.setup.ts

@@ -0,0 +1,20 @@
+/**
+ * Impide que un test filtre el codigo de salida del propio jest.
+ *
+ * Los comandos del CLI señalan su veredicto con `process.exitCode = 1` (ADR-0073:
+ * el exit code transporta el veredicto). Bajo `--runInBand` -- que es como corre
+ * CI -- no hay workers: todas las suites comparten el proceso de jest, asi que
+ * ese `1` sobrevive al test que lo puso y jest termina en 1 con 1250 pruebas en
+ * verde y la cobertura por encima del umbral. En modo worker el mismo test pasa
+ * inadvertido, y por eso el defecto solo aparecia en CI.
+ *
+ * El valor se restaura despues de CADA test. Que un test unitario fije el codigo
+ * de salida del runner es siempre un artefacto, nunca una señal: los fallos
+ * reales los reporta jest por su cuenta. Se resuelve aqui, y no en las dos specs
+ * que hoy filtran, para que la clase entera de fuga deje de ser posible.
+ */
+const original = process.exitCode;
+
+afterEach(() => {
+  process.exitCode = original;
+});