Add unit tests for hardware CLI commands

[berendt]

Closes #2364

Adds unit tests for the hardware-related CLI command modules, following the mocking patterns established in #2193 (no Celery broker, no real cloud/NetBox access):

New test modules

• tests/unit/commands/test_redfish.py: column name normalization, per-resource-type column mappings, column filtering (case-insensitive selection, unknown-column warning), JSON data filtering, table rendering, and all take_action output paths (JSON full/filtered/empty, table dispatch per resource type, unknown resource type, no result). Verifies the task lock is consulted before the Celery task is dispatched.
• tests/unit/commands/test_stress.py: default command construction of the openstack-simple-stress wrapper, boolean flag and custom value propagation, return code passthrough, FileNotFoundError/generic error handling, setup failure guard, and cleanup in all paths.

Extended test modules

• tests/unit/commands/test_baremetal.py: _apply_metalbox_vars, a parametrized cloud-setup failure guard over all commands of the module, BaremetalList (sorting, n/a power state, query flags, NetBox device-role column, cleanup on error), BaremetalDeploy (provision-state decision matrix, instance_info refresh, validation failure, NetBox local_context_data incl. cf_inventory_hostname fallback, netplan/FRR parameters, Supermicro boot device, RAID deploy steps, config-drive failure, --all loop resilience — playbook content is asserted by parsing the actual playbook.yml handed to configdrive.pack), BaremetalDump (Ironic and NetBox sources), BaremetalUndeploy (state matrix, SSH known_hosts cleanup), BaremetalPing (_ping_host output parsing incl. macOS/Linux variants, device discovery/filtering, summary), BaremetalBurnIn (clean/service step construction, manage flow, active-node confirmation), BaremetalClean (RAID step prepending, manage flow), the provide/maintenance/power wrappers, and BaremetalDelete (port-before-node ordering, NetBox state clearing on primary and secondary instances, failure resilience).
• tests/unit/commands/test_server.py: ServerMigrate (defaults, --target/--force, prompt handling, wait loop, --no-wait, PAUSED), ServerList (domain/project/user happy paths, domain-name fallback, age-based BUILD/ERROR reporting), and ServerClean (build timeout handling, prompt handling, ERROR servers, cleanup in finally).

No existing test cases were duplicated; flake8 and black pass locally.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MRk7qVcmoKuk74USmgY8Mm

[ideaship]

This asserts rc is None after the command's primary operation — playbook generation — fails inside a caught except (osism/commands/baremetal.py:564-566: log the error, then fall through with no return). cliff turns that None into exit code 0 (return_code = self.take_action(...) or 0), so a failed generation reports success — unsafe in set -e scripts and && chains.

This is the behavior the "Return non-zero on precondition and operational failures" change (Related-Bug: #2314) set out to eliminate; it fixed a subset and explicitly left the remaining baremetal operational/not-found failures open, and this is one of them. The same command already returns 1 for NetBox-unavailable and device-not-found (baremetal.py:574, 590), so the generation path exiting 0 is an internal inconsistency.

Please fix the production paths to return 1 on primary-operation failure — in their own commit — then assert exit 1 here. The same applies to the sibling assertions at lines 690, 1204, 1368, 1454, 1476, 1512, and 1667. For the batch commands (e.g. the --all delete at 1667), keep processing the remaining nodes but retain a failure flag and return non-zero at the end, so a partially-failed run doesn't report success — assert both the continuation and the final exit 1.

[ideaship]

This checks only that the private helper returns ([], []); it never exercises what take_action does with that result. In the JSON branch (osism/commands/redfish.py:175) an empty data_keys flows into _filter_json_data(result, []), which treats an empty key list as "no filtering" (redfish.py:88) and returns the full records — so --format json --column bogus on a known resource type silently prints every field, while the table path warns and prints "No valid columns specified" (redfish.py:108). The two formats disagree, and the JSON behavior has no coverage.

Please add a take_action test for --format json with an all-invalid column selection that asserts the intended behavior. Aligning the JSON path with the table path is a production fix that belongs in its own commit.