[Mohith26]

WsClient keeps a local copy of each subscribed order book, but it had three problems:

• New price levels were appended to the end of asks/bids, so after a level appears inside the book, bids[0] / asks[0] are no longer the best prices.
• The book's nonce / offset stayed at the snapshot values forever.
• Nothing checked continuity. The websocket docs say to verify that each update's begin_nonce equals the previous update's nonce (this is also what Question: How to ensure local orderbook consistency after offsets became non-contiguous? #105 asked about), but the client applied deltas regardless, so a dropped update left a silently wrong book.

This PR:

• keeps both sides sorted (asks ascending, bids descending) and applies deltas with bisect over a parallel price index instead of scanning and copying the list for every delta
• updates nonce, begin_nonce, offset, last_updated_at on the local book from each applied update
• checks begin_nonce == previous nonce; on a mismatch it stops applying deltas for that market, sends unsubscribe + subscribe to get a fresh snapshot, and calls a new optional on_order_book_gap(market_id, expected, received) callback. The server's unsubscribed ack for that resync is not reported as an unhandled message. verify_order_book_continuity=False turns the check off.
• works on both run() and run_async() (the async path awaits the resubscribe sends)
• leaves update_orders() unchanged for anyone calling it directly

Testing

• test/test_ws_client_order_book.py: 11 new tests (sorting, removals, metadata, gap -> resubscribe -> fresh snapshot, async path, disabling the check, and a 3,000-update randomized comparison against a reference book). pytest --ignore-glob="*api.py" passes, flake8 with the CI flags is clean.
• Recorded the public testnet stream for 4 markets: 30 updates, 0 continuity breaks, and for the 3 markets whose fresh snapshot had the same nonce as the last streamed update, the local book matched the snapshot exactly.
• Live on testnet: simulated a gap on market 4096; the client resubscribed, got a fresh snapshot, and kept applying the following updates with continuity intact.

Benchmark (python test/bench_ws_order_book.py, 3,000 updates of 1-8 level changes, median of 5 runs, Apple Silicon arm64):

levels per side	before	after	speedup
50	0.0375 s	0.0086 s	4.3x
200	0.1508 s	0.0098 s	15.3x
1000	0.9628 s	0.0540 s	17.8x