Fix batch progress state misalignment when resuming from mid-epoch checkpoints

[DAKSH-coder26]

Fix BatchProgress Inconsistency When Resuming From Mid-Epoch Checkpoint (Fixes #19367)

What does this PR do?

This PR fixes incorrect synchronization of batch_progress when resuming training from a mid-epoch checkpoint, as described in issue #19367.

Before this fix, resuming from a checkpoint saved before completing an epoch caused:

• batch_progress.total_ready and total_completed to desynchronize from the restored global_step
• current_ready / current_completed to become inconsistent or invalid
• Validation, checkpoint intervals, or internal loop schedules to fire at the wrong steps
• Off-by-one errors in resumed training

Summary of the Fix

The PR modifies CheckpointConnector.resume_start() to:

• Realign batch_progress.total_ready and total_completed to be at least the restored global_step
• Recompute current_ready and current_completed using
  global_step % limit_train_batches when available
• Fall back safely when dataloader size or fields differ across Lightning versions
• Ensure behavior remains unchanged when no checkpoint is provided

A new test verifies correct behavior and prevents regressions.

Tests Added

A new file:
tests/test_resume_batch_progress.py

This test:

1. Trains to global_step = 5
2. Saves a mid-epoch checkpoint
3. Resumes from the checkpoint
4. Ensures:
   • batch_progress exists in the expected form
   • total_completed ≥ global_step
   • current_completed ≤ total_completed
   • All counters are non-negative and consistent

The test failed before the fix and passes after it.

Fixes #19367

---

Before submitting

• Was this discussed/agreed via a GitHub issue? (Issue Potential off by 1 error when resuming training of mid-epoch checkpoint #19367)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing?
• No documentation changes were needed
• Added the necessary tests
• Verified new and existing tests pass locally
• Listed breaking changes (none introduced)
• Updated the CHANGELOG

---

PR review

Anyone in the community is welcome to review the PR.

Reviewer checklist

• Is this pull request ready for review?
• Check that all items from Before submitting are resolved
• Ensure the title and description clearly explain the PR
• Apply appropriate labels and milestones

---

📚 Documentation preview 📚: https://pytorch-lightning--21411.org.readthedocs.build/en/21411/

[DAKSH-coder26]

Hi maintainers , this PR is ready for review from my side.

Some required checks (schema, required-jobs) are stuck in "Expected" due to the old Lit CI platform being shut down.
Could you please trigger / approve the workflows or override the stuck checks?

Thanks!