fix: avoid duplicate originals in augment_text_with_ids

[emecii]

What does this PR do?

Summary

Fixes Augmenter.augment_text_with_ids() returning every original input twice. The method appended the source text before augmentation and then appended it again alongside the generated variants, which silently over-represented originals in the returned training data.

Additions

• Adds regression coverage for two inputs and multiple generated variants, including ID alignment.

Changes

• Retains each original input once, followed by only its generated augmentations.

Deletions

• Removes the duplicate original entry and a no-op list.extend attribute access.

Validation

• pytest tests/test_augment_api.py -k 'not back_translation and not back_transcription' -q — 12 passed, 2 model-dependent tests deselected.
• black --check, isort --check-only, and flake8 on the changed files — passed.

AI assistance

This PR was prepared with AI assistance. The behavior and regression test above were verified locally.

Checklist

• The title of this pull request summarizes its contribution.
• This description explains the changed behavior and rationale.
• Relevant regression coverage was added.
• The full existing test suite was run locally.