Fix: TCP connections never send a keepalive, causing periodic disconnects

[Fimeg]

Problem

Cameras connected over TCP (the common case — address = "ip:9000", as opposed to uid = "..." over UDP) silently drop their Baichuan session after a few minutes of otherwise-idle traffic, then reconnect. The reconnect is automatic, but it surfaces as a periodic blip in downstream consumers (Frigate, etc.), which is disruptive for continuous recording and motion detection.

Root cause

UDP sessions are kept alive by the camera itself: it sends MSG_ID_UDP_KEEP_ALIVE, which neolink acknowledges in BcCamera::keepalive(). TCP has no equivalent — nothing flows during idle periods — so the camera times the session out. There is a BcCamera::ping() (MSG_ID_PING) that looks purpose-built for this, but it is currently dead code (no call sites).

Fix

Send a fire-and-forget MSG_ID_PING every 60s on TCP connections only, to hold the session open. UDP is left untouched, since it already keeps itself alive.

• BcCamera::new() now tracks whether the transport is UDP and, for TCP, spawns a task that sends MSG_ID_PING on an interval. The task ends on its own once the connection goes away (the send returns an error).
• BcConnection::send is widened from pub(super) to pub(crate) so the keepalive can be sent without subscribing for a reply.

Validation

The same approach (a periodic MSG_ID_PING on TCP) has been running in production on TCP/Baichuan cameras (port 9000) where it eliminated reconnects that previously recurred every few minutes without fail. This PR ports that fix to current master.