chore(api): drop gating from the allow_zero_dpu_hosts flag

[chet]

Description

This rips out the allow_zero_dpu_hosts flag entirely. With per-host and site-wide dpu_mode available, we shouldn't need to say "allow zero DPU hosts" -- we should simply express what ExpectedMachines are expected to be DpuMode::NoDpu (or ::NicMode, or ::DpuMode), which we can also set at the site level now too.

For example we explicitly say:

• If an ExpectedMachine is marked DpuMode::DpuMode (per-host or site-level default), we expect DPUs; if site-explorer doesn't find any, it logs a warning and skips ingestion.
• If an ExpectedMachine is marked as DpuMode::NoDpu, we DON'T expect DPUs; if the BMC reports some, site-explorer ignores them and ingests the host as zero-DPU.
• If an ExpectedMachine is marked as DpuMode::NicMode, site-explorer effectively ingests the host as zero-DPU; when it explores a NicMode host with DPUs, it will set_nic_mode on them as needed, naturally removing them from the DPU list across exploration cycles until all DPUs are reporting NicMode.

...no need for saying allow_zero_dpu_hosts. If an ExpectedMachine isn't marked NoDpu, then we don't expect it to be. If it is, then we do.

Any configs out there that still have allow_zero_dpu_hosts = true will keep parsing fine -- it just won't do anything going forward; setting dpu_mode: no_dpu or dpu_mode: nic_mode (or dpu_mode: dpu_mode) is all that needs to be set on the ExpectedMachine (or site-level config).

Went through existing tests to make sure we've got good coverage (and added some). Table below:

Case	Action	Coverage
Zero-DPU host w/ ::NoDpu set	ingest as zero-DPU	test_site_explorer_ingests_no_dpu_host (and test_site_explorer_main)
Zero-DPU host, w/ ::DpuMode set	skip/warn/metric	test_site_explorer_skips_unexpected_zero_dpu_host
Zero-DPU host w/ ::NicMode set	ingest as zero-DPU	test_site_explorer_ingests_nic_mode_host_with_no_observed_dpus
NicMode set w/ DPU still in DPU mode	fix via set_nic_mode	test_site_explorer_auto_corrects_nic_mode_per_expected_machine
Err(NoDpu) + expected_dpu_count == 0	return Ok(None)	handle_no_dpu_error_treats_as_ok_on_zero_dpu_host
Err(NoDpu) + expected_dpu_count > 0	return the error	handle_no_dpu_error_surfaces_when_dpus_were_expected
BMC call succeeded (any expected_dpu_count)	return the success untouched	handle_no_dpu_error_passes_through_success
BMC failed for a non-NoDpu reason (e.g. lockdown)	return the error untouched (only NoDpu is ever swallowed)	handle_no_dpu_error_does_not_touch_other_errors
Integration e2e for Zero-DPU (machine-a-tron)	(None)	test_machine_a_tron_singledpu_nic_mode + integration tests

Signed-off-by: Chet Nichols III chetn@nvidia.com

Type of Change

• Add - New feature or capability
• Change - Changes in existing functionality
• Fix - Bug fixes
• Remove - Removed features or deprecated functionality
• Internal - Internal changes (refactoring, tests, docs, etc.)

Related Issues (Optional)

Breaking Changes

• This PR contains breaking changes

Technically it just makes it so dpu_mode: no_dpu works now without needing to also set allow_zero_dpu_hosts: true.

Testing

• Unit tests added/updated
• Integration tests added/updated
• Manual testing performed
• No testing required (docs, internal refactor, etc.)

Additional Notes