Database errors should have their own status_code

[evansd]

We currently append some custom text to the error message when a database job fails with certain exit codes, but we leave the status_code as NONZERO_EXIT like any other failed job. This makes monitoring and debugging harder than it needs to be. We should add custom status codes for this. I think probably one for each specific type of database error?

Related Slack thread:
https://bennettoxford.slack.com/archives/C069YDR4NCA/p1736417917622859

And relevant bits of code:

job-runner/jobrunner/run.py

Lines 448 to 451 in 4fba743

	elif job_definition.allow_database_access:
	error_msg = config.DATABASE_EXIT_CODES.get(results.exit_code)
	if error_msg:
	message += f": {error_msg}"

job-runner/jobrunner/config.py

Lines 218 to 227 in 4fba743

	DATABASE_EXIT_CODES = {
	# Custom database-related exit codes return from cohortextractor, see
	# https://github.com/opensafely-core/cohort-extractor/blob/0a314a909817dbcc48907643e0b6eeff319337db/cohortextractor/cohortextractor.py#L787
	3: (
	"A transient database error occurred, your job may run "
	"if you try it again, if it keeps failing then contact tech support"
	),
	4: "New data is being imported into the database, please try again in a few hours",
	5: "Something went wrong with the database, please contact tech support",
	}

job-runner/jobrunner/models.py

Lines 69 to 76 in 4fba743

	# FAILED states
	DEPENDENCY_FAILED = "dependency_failed"
	NONZERO_EXIT = "nonzero_exit"
	CANCELLED_BY_USER = "cancelled_by_user"
	UNMATCHED_PATTERNS = "unmatched_patterns"
	INTERNAL_ERROR = "internal_error"
	KILLED_BY_ADMIN = "killed_by_admin"
	STALE_CODELISTS = "stale_codelists"