fix(atlas): Preserve preferred affinity fallback during resize

[ryzenate1]

Issue

A moving resize with placement rules can skip the Preferred affinity fallback.

When the current Metal Server cannot fit the resize delta and no affinity-matching destination can fit the target shape, a nonmatching server with enough capacity should remain available under Preferred.

The affinity filter currently returns as soon as a resize has a current host, before reading the Affinity Matching setting. This removes the nonmatching fallback and the resize ends in OutOfCapacity.

For example:

• A matches the placement rule but cannot fit the target.
• B does not match the rule but can fit the target.
• C is the current host but cannot fit the resize delta.
• Affinity Matching is Preferred.

Before this change, placement keeps A and C, drops B, and returns OutOfCapacity.

Summary

Evaluate the affinity matching mode when none of the retained current/matching hosts has room.

Preferred can then fall back to the full host set, while Enforced continues to keep the current/matching set and never uses a nonmatching destination.

Why

A resize always has a current host. The previous early return treated the existence of that host as sufficient reason to keep the affinity-filtered snapshot, even when the current host could not accommodate the resize.

That made the Preferred branch unreachable for this moving-resize case.

This keeps the current host in the candidate set for in-place resize, but only commits to the affinity-filtered set when one of those candidates actually has room.

What changed

• Let a Preferred moving resize fall back when neither the current host nor an affinity-matching host has capacity.
• Preserve the restricted candidate set for Enforced.
• Add a regression test covering a full current host, a full matching host, and a fitting nonmatching fallback.

This preserves the Preferred fallback semantics introduced with the affinity placement behavior in #310 for moving resizes.

Validation

• Regression test fails on the pristine implementation (Retained hosts: ['a', 'c']; affinity setting reads: 0) and passes with this change.
• Verified current-host delta-fit still selects the current host.
• Verified a fitting affinity-matching destination is still preferred over a nonmatching host.
• Verified Enforced never falls back to the nonmatching host.
• Verified Preferred VM creation without a current host still falls back as before.
• Verified contention still returns PlacementBusy instead of relaxing affinity.
• Placement context suite: 48 tests run, 0 failures, 8 pre-existing DB-integration errors unchanged from baseline; all 4 key ATLAS-C1 controls GREEN.
• Reproduced on real AWS/Metal infrastructure with an isolated temporary Atlas site and three independent KVM Metal guests with real metald sync and real capacity reports: pristine placement returned OutOfCapacity; patched placement selected the nonmatching fallback. No production site or customer workload was modified.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[tanmoysrt]

Try to keep comments, docstrings smaller and easy to understand. Trimmed it for now.

[ryzenate1]

Thanks, noted. I’ll keep comments and docstrings concise and limited to non obvious behavior going forward 😃