[feature request]: Ability to reattach live-remote viewers

### Is there an existing proposal for this?

- [x] I have searched the existing proposals

### Is your feature request related to a problem?

I'm doing some long-term monitoring of server processes to try and find a leak. I do this by having the server run under memray --live-remote, and looking at it in a memray live tui from an ssh session. If the session times out, I have to end my inspection, because there's no way to reattach.

### Describe the solution you'd like

I would love if memray, when running in `--live-remote`, would
- (ideally) continue to block on start until a viewer connects, though I think it would also be fine to say "start the remote tui before starting the process under inspection"
- upon disconnection of the viewer, reopen a listen socket and start dropping data until something disconnects
- since this is a live view, it seems okay to do this - obviously wouldn't want to for any data aggregation modes

### Alternatives you considered

Having temporal logging also seems like it would work okay so you could inspect it after the fact, but I can see those getting really big - having a live view that's up to the user to inspect at whatever frequency they want is a perfect solution to these kinds of slow leaks in programs with many different internal lifecycles.

Start by tracing the --live-remote entry point and how the live TUI connects to the server over its listen socket. Define the reconnect behavior for viewer disconnects, including whether startup waits for a viewer and how data is handled while none is connected. Done means a disconnected viewer can reconnect to the same live process without stopping inspection.

Written by the indexing model from the issue text.