perf: add benchmark for ChromaDB retrieval latency (#311)

[Aharshi3614]

What this does

Adds a benchmark test suite for ChromaDB retrieval latency in backend/tests/test_chromadb_benchmark.py.

Why

Closes #311

Benchmark Results

| Test | Result |
| Single query latency | 11.886ms |
| Mean over 10 runs | 20.248ms |
| Small collection (10 chunks) | 17.206ms |
| Large collection (200 chunks) | 19.899ms |
| Empty collection | 7.822ms |
| Latency stdev (consistency) | 1.922ms |

Test Coverage

• Single query latency
• Repeated query latency (10 runs with stats)
• Small collection (10 chunks)
• Large collection (200 chunks)
• top_k variation (1, 4, 8)
• Empty collection handling
• Latency consistency check

Testing

pytest tests/test_chromadb_benchmark.py -v -s
7 passed in 21.33s

[imDarshanGK]

@Aharshi3614 Thanks for the contribution. The PR is generally aligned with Issue #311, but I noticed a few things that need to be addressed before it can be merged:

1. In test_single_query_latency, the _retrieve() arguments appear to be reversed.

Current code:
context, docs = _retrieve("document information topic", session_id, top_k=4)

But _retrieve() is defined as:
def _retrieve(session_id: str, query: str, top_k: int = 4):

It should be:
context, docs = _retrieve(session_id, "document information topic", top_k=4

2. The collection cleanup should use the same collection naming logic as creation to avoid potential mismatches in the future.

3. If possible, avoid loading SentenceTransformer at module import time, since it can slow down test collection and CI runs.

Please update the PR and let me know once the changes are pushed.