dob: scale wire qty fields to per-instrument qty_exponent

[armcconnell]

Summary of Changes

The DoB emitter was writing OrderAdd.quantity, OrderExecute.exec_quantity, and SnapshotOrder.quantity at the publisher's internal 10^8 fixed-point (Sz::value()) while the accompanying InstrumentDefinition reported the venue's per-instrument qty_exponent. Subscribers decode display = raw * 10^qty_exponent per spec, so the on-wire error was 10^(8 + qty_exponent)x for every instrument with qty_exponent != -8 — i.e. every HL instrument. TOB was correct because it parsed the original string with qty_to_fixed(level.sz(), inst.qty_exponent); this PR makes DoB do the equivalent.

• order_book::sz_to_fixed: helper that converts an internal Sz (qty * 1e8) to the wire u64 for a given qty_exponent. Lossless for HL: venue qty is always a multiple of 10^qty_exponent, so the divisor 10^(8 + qty_exponent) divides exactly.
• dob_tap: CoinResolver widened to Fn(&Coin) -> Option<(u32, i8)> so the apply tap can scale per-instrument. Applied at OrderAdd.quantity and OrderExecute.exec_quantity emission.
• multicast/dob: emit_snapshot takes qty_exponent and applies the same scaling at SnapshotOrder.quantity. run_dob_snapshot_task threads qty_exponent through the priority and round-robin paths from the registry.
• websocket_server: construction-site resolver returns qty_exponent alongside instrument_id from the registry's active map.

Testing Verification

• cargo test --workspace: 119/119 lib + tests pass on the branch (the single pre-existing test_trade_listener failure reproduces on main un-modified and is unrelated). cargo build --release succeeds. cargo clippy --lib --tests --no-deps produces no errors.
• New unit tests in dob_tap.rs::tests assert OrderAdd.quantity and OrderExecute.exec_quantity scale correctly at qty_exponent = -3 and qty_exponent = 0 (the integer-only case from the live observation).
• New multicast::dob::snapshot_qty_scaling_tests decode the actual SnapshotOrder bytes off a UDP loopback at qty_exponent = -3 and qty_exponent = 0 and assert the wire quantity field matches the venue scale.
• parity_tests.rs end-state TOB-vs-DoB scenario factored to take qty_exponent and instantiated at -8, -3, and 0. The -3 and 0 variants would have failed pre-fix with dob_qty / tob_qty == 10^(8 + qty_exponent), exactly matching the symptom from the field.
• Deployed to aws-tyo-hl-mainnet via the ansible role with dz_hl_publisher_branch=fix/dob-qty-exponent; service restarted, publisher healthy, on-wire output verified as expected.

Refs: hyperliquid hyperliquid-dex#10