refactored architecture for more programatic usage

[pavanjava]

No description provided.

[srimon12]

Findings:

1. run_query() is documented as a one-shot helper that closes the connection, but it never actually closes the Connection, so repeated programmatic calls will leak QdrantClient instances. In src/qql/init.py the docstring says it “creates a Connection, runs one query, and closes the connection,” but the implementation at src/qql/init.py just returns Connection(...).run_query(query). I verified this with a small mock probe: run_query() called conn.run_query() once and conn.close() zero times. The new tests in tests/test_connection.py only assert delegation and also miss the close call entirely. Using with Connection(...) as conn: inside run_query() or explicitly calling close() after execution would fix both the behavior and the doc mismatch.

2. The README’s test-count documentation is inconsistent inside the same PR. The badge at README.md says 549 tests, and docs/reference.md was updated to 549, but the README still says “Expected: 500 tests passing” at README.md. I re-ran collection on the PR worktree and it reports 549 tests collected, so the README footer is stale.

I did not find other substantive code/test/doc gaps in this PR beyond those two. The new Connection surface itself is otherwise covered reasonably: 549 tests collect successfully on the PR snapshot, and tests/test_connection.py passes cleanly.