[Feature] Built-in collection versioning for zero-downtime index updates

## Description

When re-indexing a collection with updated embeddings or new documents, having built-in collection versioning would enable atomic swaps between old and new indexes without any query downtime.

## Proposed Pattern

```python
# Build new version in background
v2 = client.create_collection('docs_v2')
# ... populate v2 ...
# Atomic swap
client.swap_alias('docs', 'docs_v2')
```

This is a common pattern in search infrastructure (Elasticsearch aliases, Qdrant collection aliases) and would be very useful here.

Start by reviewing the existing collection lifecycle and client API around create_collection; the issue names no files or tests. Compare the proposed create_collection('docs_v2') and swap_alias('docs','docs_v2') flow with current behavior, then clarify atomicity, version visibility, and completion tests with maintainers.

Written by the indexing model from the issue text.