1.9.0 — Aqualung/Oceanic/Sherwood BLE fixes

Fixes Oceanic/Aqualung/Sherwood BLE downloads (#24)

Aqualung/Oceanic/Sherwood devices (Atom2 protocol family — i300C, i330R, i550C, i770R, i200C, and rebrands) were completely unable to connect and download over BLE. This release fixes the full chain of issues, confirmed working end-to-end against a real Aqualung i300C.

Fixed

• DC_IOCTL_BLE_GET_NAME was a stub returning DC_STATUS_UNSUPPORTED. oceanic_atom2_device_open() needs the peripheral's advertised BLE name for its handshake; without it the device NAKed every command.
• Characteristic selection could bind to the wrong write/notify characteristic. Devices exposing more than one write- or notify-capable characteristic in the same service (the i300C has a data channel plus an auth-nonce channel) could end up bound to the wrong one depending on BLE enumeration order, producing "Unknown ATT error" on write.
• No retry on dc_device_open. The i300C (and other Pelagic OEMs) ignores protocol commands for several seconds after a fresh BLE link is established. Now retries up to 5 times with a 3s delay on the same link, instead of failing immediately or forcing an expensive disconnect/reconnect cycle.
• connect(toDevice:) didn't wait for a prior connection to fully tear down. CoreBluetooth reuses CBPeripheral instances; reconnecting while a previous teardown was still in flight triggered internal state-machine errors that delayed or blocked the retry. Now waits up to 5s for .disconnected before issuing a new connect.
• Fingerprint lookup returned a hardcoded 4-byte sentinel when nothing was stored. Oceanic/Aqualung's Atom2 family needs 8-byte fingerprints; the wrongly-sized buffer made dc_device_set_fingerprint fail silently, so fingerprinting never took effect on these devices — every sync re-downloaded all dives instead of just new ones. Now returns nil/0, which the bridge already knows how to skip safely.
• close_device_data() leaked the device descriptor on every close, for every device family, not just Aqualung — it nulled the pointer without freeing it. Two other failure paths in open_ble_device() had the same gap.
• Redundant reconnect attempts. open_ble_device_with_identification() could retry the same protocol family twice (once via stored config, once via name-based fallback) when both would fail the same way — now skips the redundant attempt.

Also included

• Dropped a stale TODO about the Shearwater Perdix 3 descriptor table entry — resolved upstream in libdivecomputer, Perdix 3 already works.

Credits

Several of these fixes (dc_device_open retry, the connect(toDevice:) disconnect-wait, and the fingerprint sentinel fix) were found and verified working by @houle988 against a real device, and ported from his fork. Thanks to @kreitje for the original report and patient retesting through each iteration.