[CircleCI] Enable CI for private forks, no-op for public repo (#48056)

* [CircleCI] Enable CI for private forks, no-op for public repo

- Replace no-op config.yml with full CircleCI setup using
  `pipeline.project.git_url` to differentiate public vs private repos:
  - Public `huggingface/transformers`: genuine noop workflow
  - Private fork: full CI (check_circleci_user, check_code_quality,
    check_repository_consistency, fetch_tests with TRANSFORMERS_CONTEXT)
- Add `[run-circleci]` gate as a reusable `commands` block to control
  CI cost — jobs halt early unless commit message contains `[run-circleci]`
- Fix resource_class default: xlarge → large
- Fix parallelism expression YAML quoting in generated config
- Halve pytest workers to fit large resource class: -n workers//2
- Add peft_integration CircleCI job to match tests_fetcher.py routing

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [CircleCI] Remove examples_torch job — too large for large runner

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [CircleCI] Remove examples_torch from tests_fetcher.py to avoid unknown parameter error

Without this, tests_fetcher.py would still generate examples_torch_test_list.txt,
process_test_artifacts.py would include examples_torch_parallelism in the continuation
parameters, and CircleCI would reject with an unknown pipeline parameter error since
examples_torch is no longer in ALL_TESTS.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Revert "[CircleCI] Remove examples_torch from tests_fetcher.py to avoid unknown parameter error"

This reverts commit b2cc4841b7c7073151e3fda03c2356322aa03404.

* [CircleCI] Add disabled flag to CircleCIJob; disable examples_torch

Adds a `disabled` field to `CircleCIJob`. When True, `to_dict()` returns
a minimal noop job (cimg/base:stable, small runner, single echo step)
instead of the full test job. The job stays in ALL_TESTS so pipeline
parameters are still declared and the continuation doesn't fail with
unknown parameter errors.

Mark examples_torch as disabled: the examples-torch docker image is too
large for the large resource class and would cause OOM crashes.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>