Fix Hub download filtering for FlashPack pipelines

[rumutaydin]

What does this PR do?

Pipelines saved with save_pretrained(use_flashpack=True) currently cannot be loaded
back through the Hub. Such a pipeline stores diffusers models as model.flashpack and
transformers components (text encoders) as safetensors but the download path breaks
on this layout three ways:

1. _get_ignore_patterns takes the elif use_flashpack: branch and ignores all
   *.safetensors. Since use_flashpack is only forwarded to diffusers models, text
   encoders can only load safetensors which are now missing from the snapshot:

   from diffusers import AutoPipelineForText2Image
   
   pipe = AutoPipelineForText2Image.from_pretrained("hf-internal-testing/tiny-flux-pipe")
   pipe.save_pretrained("flashpack-repo", use_flashpack=True)   # + push to Hub
   AutoPipelineForText2Image.from_pretrained("<user>/flashpack-repo", use_flashpack=True)
   # OSError: Error no file named model.safetensors, or pytorch_model.bin,
   #          found in directory .../text_encoder

   With explicit use_safetensors=True it raises EnvironmentError instead, because
   the flashpack-only folders make is_safetensors_compatible fail.

2. FlashPack files are missing from variant_compatible_siblings' weight names, so
   they never enter the download allow-patterns combined with (1), a FlashPack
   pipeline snapshot contains no weights at all.

3. use_flashpack=False: the download docstring says FlashPack weights "will never
   be downloaded", but *.flashpack appears in no ignore pattern, so default users of
   dual-format repos download FlashPack blobs they never read.

Fixes (issue)

judge safetensors compatibility over non-flashpack folders only, keep just the
flashpack file inside flashpack-covered folders (per-folder ignore patterns), register
FLASHPACK_WEIGHTS_NAME as a weight name, ignore *.flashpack whenever
use_flashpack=False, and forward use_flashpack from from_pretrained to download
(it was popped but never passed on, so the flag had no effect on Hub downloads).

Repos with no FlashPack files behave exactly as before, with or without the flag.

Tests

an end-to-end regression test that saves a FlashPack pipeline, applies the real
allow/ignore download filter, and round-trips the filtered snapshot (fails on main),
plus direct _get_ignore_patterns cases for the mixed/dual/flashpack-only/no-flashpack
layouts and a test that from_pretrained forwards the flag.

Related:

#12564, #12700.

Before submitting

• [x ] Did you use an AI agent (Claude Code, Codex, Cursor, etc.) to help with this PR? If so:
  • [x ] Did you read the Coding with AI agents guide?
  • [x ] Did you self-review the diff against .ai/review-rules.md?
• [ x] Did you read the contributor guideline?
• [x ] Did you read our philosophy doc? (important for complex PRs)
• Was this discussed/approved via a GitHub issue or the forum? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• [x ] Did you write any new necessary tests?
• Are you the author (or part of the team) of the model/pipeline (only applicable for model/pipeline related PRs)?

Who can review?

@DN6 @yiyixuxu

[github-actions]

Hi @rumutaydin, thanks for the PR! It does not appear to link an issue it fixes. If this PR addresses an existing issue, please add a closing keyword (e.g. Fixes #1234) to the PR description so the issue is linked. See the contribution guide for more details. If this PR intentionally does not fix a tracked issue, a maintainer can add the no-issue-needed label to silence this reminder.