[LoRA] apply .alpha to Qwen-Image lora_A/lora_B LoRAs

[dcaayushd]

What does this PR do?

Fixes #14938

Qwen-Image LoRAs using lora_A/lora_B weights could lose their per-module .alpha values during conversion. When alpha differed from the LoRA rank, the adapter was loaded at the wrong strength—for example, a rank-32 LoRA with alpha 83 loaded at approximately 1/2.6 of its intended strength.

This change:

• Applies .alpha / rank during lora_A/lora_B conversion.
• Raises a clear ValueError when a matching lora_A or lora_B weight is missing.
• Removes leftover .alpha metadata.
• Preserves tensor dtype and device without casting.
• Keeps the existing lora_down/lora_up conversion behavior unchanged.

Added tests cover alpha values of None, 2, 4, and 8, unmatched A/B weights, stray alpha keys, and bfloat16 dtype preservation.

Real-checkpoint verification with e-n-v-y/Qwen-Image-2.1-Fix-v2.0 showed:

• 30 .alpha keys in the raw checkpoint.
• 0 .alpha keys after conversion.
• The converted weights match (alpha / rank) * (lora_B @ lora_A) with maximum absolute differences between 5.6e-09 and 3.1e-05 across three sampled modules.

The existing lora_down/lora_up conversion output was identical to the upstream output for a synthetic Qwen-Image-Lightning checkpoint.

Before submitting

• Did you use an AI agent (Claude Code, Codex, Cursor, etc.) to help with this PR? If so:
  • Did you read the Coding with AI agents guide?
  • Did you run the self-review skill? The skill was not installed in the environment; its checked-in rubric was applied manually instead.
  • Did you share the final self-review notes in the PR description or a comment?
• Did you read the contributor guideline?
• Did you read our philosophy doc? This is a focused LoRA conversion fix.
• Was this discussed/approved via a GitHub issue or the forum?
• Did you make sure to update the documentation with your changes? No user-facing documentation changes are needed for this converter fix.
• Did you write any new necessary tests?
• Are you the author (or part of the team) of the model/pipeline?

Validation

• Focused new tests: 8 passed.
• Qwen-Image LoRA tests: 36 passed, 7 skipped; 2 unrelated MPS group-offloading failures.
• LoRA regression tests: 81 passed, 56 skipped; 4 unrelated MPS group-offloading failures.
• ruff check: passed.
• ruff format --check: passed.
• git diff --check: passed.

Self-review notes

The checked-in self-review rubric was applied manually because the self-review CLI skill was unavailable. No blocking correctness, dead-code, copied-code, or documentation issues were found.

AI assistance was used for implementation, testing, verification, and drafting this description. Please review and edit the wording before submitting.