[mhellevang]

On top of #227 (needs it). Fixes the places where LoginSession differed from the references. With the state machine each fix is a few rows in the table. One commit each:

1. Reconnect window runs from the bus reset. Apple starts the reconnect timer in suspendedNotify (management timeout + 1 s) and restarts it on every further reset (IOFireWireSBP2Login.cpp:2067-2077, 2202-2235). If the login hasn't reconnected when it expires, it's lost (reconnectTimeout, :2395-2413). We started the timer only when the reconnect write was ACKed, and waited for rediscovery without a limit. The window is now a second timer beside the management one. If it closes before rediscovery, the session reports lost and rediscovery logs in afresh. On the LS-9000 that's a ~3 s window against the 650 ms rediscovery measured in SBP-2: re-login/reconnect after rediscovery, not on a 100 ms timer #216.
2. Resend the reconnect on "unspecified error". Apple treats request complete + sbp_status 0xFF as transient and resends after 100 ms, inside the window (:2498-2503, :589). We fell back to a fresh login. Other rejections still fall back. The LS-9000's rejection after a power-cycle is sbpStatus=4, so it's unchanged.
3. Ignore a bus reset while logging out. Apple's suspendedNotify does nothing in LoggingOut (:2088-2092). We dropped to Idle silently. Now the old-generation logout write fails, or the logout timeout fires, and either one reports the logout.

Kept as is, documented: logout while suspended follows Linux (sbp2_remove logs out at once, sbp2.c:1221-1243). Apple's fLogoutPending path never actually sends one: in Reconnect it only sets the flag again, and the timeout calls executeLogout with the flag still set, which returns busy (:2409-2410, :2634).

Not tested on hardware yet. Plan for the LS-9000:

• a cable pull with a quick replug (reconnect inside the window),
• a power-cycle (rejected → login),
• unplugged for longer than the window (lost → login on replug).

An ID-0 reconnect on the H4D is still open.