Don't let a pyatv error during start up kill the child bridge

[seidnerj]

Problem

Any error raised while an accessory is starting up takes the whole child bridge down. From my log:

Error: Got pyatv Error: cannot convert float NaN to integer
    at parseState (.../@sebbo2002/node-pyatv/src/lib/tools.ts:239:15)
    at A.getState (.../@sebbo2002/node-pyatv/src/lib/device.ts:535:30)
    at AppleTVEnhancedAccessory.startUp (src/appleTVEnhancedAccessory.ts:1704:17)

[Apple TV Enhanced] Child bridge ended (code 1, signal null).
[Apple TV Enhanced] Child bridge will automatically restart in 20 seconds (restart attempt 2 of 4).

Homebridge only retries four times, so a device that is unreachable or misbehaving for a couple of
minutes leaves the accessory gone until Homebridge is restarted by hand.

Cause

validationLoop in the accessory constructor discards three promises with void:

//FIXME: catch errors / remove void
void this.credentialsValid().then((valid: boolean): void => {
    if (valid) {
        this.log.success('Credentials are still valid. Continuing ...');
        void this.startUp();
    } else {
        this.log.warn('Credentials are no longer valid. Need to repair ...');
        //FIXME: catch errors / remove void
        void pairingRequired().then(validationLoop.bind(this));
    }
});

All three talk to the device through pyatv, and startUp() in particular does a lot of it
(getPowerState, getDeviceState, getMediaType, createVolumeFan -> getState, listApps).
void throws the promise away along with its rejection, so nothing is left holding it; on Node 18+
an unhandled rejection terminates the process by default, which is the code 1 above. The two
FIXME: catch errors / remove void comments already mark this.

Change

Keep hold of the chain and attach a catch that logs the failure and retries the whole validation
loop after a delay, rather than letting the rejection escape. Both FIXMEs are addressed; the happy
path is unchanged.

Testing

npm run build passes (lint at --max-warnings=0 + tsc). Running on a Raspberry Pi with three Apple
TVs; bridge starts normally and the accessories behave as before. The retry path itself only triggers
on a pyatv failure, so it is exercised by the fault rather than by the happy path.

Note

The NaN error that led me here is a pyatv bug and I have a fix open for it separately
(postlund/pyatv#2897). This change is independent of that one: it is about any pyatv error during
start up being fatal to the child bridge rather than recoverable.