Synchronisation requires a break character, which is effectively the line being high for more than a complete character's worth of time.
A "break" character is the line being LOW for more than a complete character time. What you want for re-synchronization (and what you describe) is "line idle state." (line HIGH for a while.)
Also, it doesn't actually have to last for more than a complete character time; it just has to continue till the idle state overlaps where there should be a stop bit. (a full byte+ time guarantees this.)
Note also that if you transmit continuously on a UART interface, then eventually you'll lose characters because of the slight difference in baud rate between the two ends. There must always be occasional breaks to ensure that the long-term average data rate from the transmitter doesn't exceed the Rx rate at the receiver.
In theory, that's the "STOP bit" that is already present between characters. AFAIK, most UARTs wil report a "new" byte received as soon as they detect a STOP bit, which could be any HIGH state following the last data bit, or could be after whatever algorithm the UART uses to detect a "bit" (probably sampling about half way through the bit.) After "detecting" the stop bit, the UART will then
asynchronously wait for the next START bit.
This is why UARTs sample at (typically) 1/16 the baud rate, and what the "asynchronous" part in the name means; they do NOT require that the bits show up on bit-time clock edges.
Also in theory, if the UART detects a "framing error", it will reset the logic and start looking for a start bit (LOW state) again.
That means that with "random" continuous data will probably re-synchronize all by itself, after a while. A data pattern that
always puts a HIGH state on the line when the UART expects a stop bit (and therefore does not cause a framing error) is pretty uncommon. Many "simple" UART drivers don't bother to do anything special WRT framing errors, but the UART hardware still does ... something.
I've seen this happen in real life, with data sources like newswires connected to "Terminal Servers." Reboot the terminal server, and the data coming out the other end is wrong for some time, but ... not that long.
UART receivers implemented in software (bit-banged) don't usually over-sample their bit detection, but may tie their start-bit detection to a much faster clock (ie some implementations use a pin-change interrupt to detect the start bit, so they're pretty much sampling the data at something near the CPU clock rate.)
It would be an interesting experiment to see how short the stop bit can be before a UART doesn't like it...
Different UART implementations may behave slightly differently. But not very much. This is really old technology at this point, such that anything that behaves significantly differently would be labeled "buggy."