test(go): add unit tests for Go API

[kauereinbold]

Summary

• pkg/api/handler_test.go: all six handlers through httptest with a scripted fake repository: success, 404, 400 and 500 paths (19 cases).
• pkg/api/repository_test.go: Postgres repository through go-sqlmock (test-only dependency): GetAll, Count, GetByID, Create, Update, Delete, asserting SQL shape (deleted filter, argument order, soft delete) and error propagation (15 cases).
• Fix: Count called log.Fatal on a database error, which terminated the API process; it now returns the error and the handler answers 500.
• CI go-pull-request.yml: go test -coverpkg=./... -coverprofile plus a coverage gate (MIN_COVERAGE=70). Current total 78.3% (pkg/api 98.1%, cmd/app/main.go uncovered).

Closes #303

Test plan

• go test ./... -count=1 -coverpkg=./... green locally, 78.3% total
• go vet, gofmt -l clean; actionlint clean
• Gate verified locally: passes at 70, fails at 99
• CI go - build - pull request green with the new coverage step

[github-actions]

Reviewed the diff against CLAUDE.md/agents.md. No blockers or majors.

[minor] repository_test.go (TestRepositoryGetByID, "other errors also map to ErrorReminderNotFound"): the test pins down existing behaviour in GetByID where any DB error (not just sql.ErrNoRows) is mapped to ErrorReminderNotFound and surfaced as a 404, masking real failures as "not found". Out of scope for this PR (Count fix only), but worth a follow-up issue since the test now locks the bug in as intended behaviour.

[minor] The new coverage gate (MIN_COVERAGE=70, hard exit 1 below threshold) is a new enforced process rule; the dotnet and react PR workflows only collect/upload coverage today without failing the build. Not a blocker, but if this is meant to become the standard, worth a short ADR note or a follow-up to align the other pipelines for consistency.

[nit] Coverage gate script recomputes total via awk twice conceptually (once for parsing, once inside the awk comparison) - fine as is, just noting it's a bit dense; a comment explaining the double-negative exit !(t < m) would help future readers.

Everything else checks out: the Count() log.Fatal → returned error fix matches the PR description and handler test coverage, SQL argument order/soft-delete/route-id-over-body-id assertions in repository_test.go are correct, go-sqlmock is a test-only dependency, PR title is conventional (test(go): ...), no secrets/dashes/AI attribution found.

Ready for human review.