database: Don't allow empty code host config

[ryanslade]

We already have db constraint to stop this in most cases, but when we have encryption enabled then even an empty config is converted into an encrypted value so we need to do the check in our application layer too.

Closes https://github.com/sourcegraph/sourcegraph/issues/41924

Test plan

New unit tests added

[sourcegraph-bot]

Codenotify: Notifying subscribers in CODENOTIFY files for diff 26a1f95159ab8e353c2c892b09d4b62b70dd0ea5...a1e91f4b24f8584f8be840e5516f61be76c9a9f8.

Notify	File(s)
@eseliger	enterprise/internal/batches/testing/repos.go internal/database/external_services.go internal/database/external_services_test.go internal/database/repos_perm_test.go internal/extsvc/types.go internal/usagestats/batches_test.go
@indradhanush	enterprise/cmd/repo-updater/internal/authz/integration_test.go internal/repos/github_webhook_handler_test.go internal/repos/status_messages_test.go internal/repos/store_test.go internal/repos/sync_worker_test.go internal/repos/syncer_test.go internal/repos/webhook_build_handler_test.go
@sashaostrikov	internal/repos/github_webhook_handler_test.go internal/repos/status_messages_test.go internal/repos/store_test.go internal/repos/sync_worker_test.go internal/repos/syncer_test.go internal/repos/webhook_build_handler_test.go
@unknwon	enterprise/cmd/repo-updater/internal/authz/integration_test.go internal/database/external_services.go internal/database/external_services_test.go internal/database/repos_perm_test.go internal/repos/store_test.go

[eseliger]

Should we instead say the config has to be parseable?

[ryanslade]

Should we instead say the config has to be parseable?

@eseliger I'm going to look into this today.

We already have the MakeValidateExternalServiceConfigFunc function which appears to do what we want and even checks the JSON against our schema. I'll look into whether we've missed calling that somewhere or if there's a bug somewhere in that code.

[eseliger]

Might be that this PR would fix it, too (didn't dive too deep): https://github.com/sourcegraph/sourcegraph/pull/42039

[ryanslade]

Might be that this PR would fix it, too (didn't dive too deep): #42039

That PR assumes we were doing the validation correctly in all cases, but we weren't doing it for the Upsert case.

[ryanslade]

PTAL

[ryanslade]

This change has broken a lot of tests outside of the internal/database package. I'll go though and update them. I think in most cases it's because we use dummy code host config that doesn't conform to our schema.