Timestamp in events

Nobody has claimed this yet.

Events don't have a timestamp, but they should. The main reason they don't yet is that it's a tricky problem to choose an appropriate type for the timestamp and convert timestamps from the system into it. Should this timestamp be comparable to [`std::time::Instant`](https://doc.rust-lang.org/std/time/struct.Instant.html)?

On Windows, there is a `time` field in the [`MSG`](https://docs.microsoft.com/en-us/windows/win32/api/winuser/ns-winuser-msg) struct. This is helpfully documented as "The time at which the message was posted." Based on a [Raymond Chen blog](https://devblogs.microsoft.com/oldnewthing/20140122-00/?p=2013), I believe this is 32 bit time in milliseconds as returned by [GetTickCount](https://docs.microsoft.com/en-us/windows/win32/api/sysinfoapi/nf-sysinfoapi-gettickcount), and also with very poor resolution guarantees (in other words, it may or may not be a lower quality time source than calling `Instant::now()` in the window message handler). The 32 bitness is also a problem, it will overflow in a couple of months.

On macOS, it's [NSEvent.timestamp](https://developer.apple.com/documentation/appkit/nsevent/1528239-timestamp?language=objc), which is an interval since system startup, as a 64 bit float. Again, this cannot be compared to an `Instant::now()`, for example to measure the delay between when the event was created and when it was processed.

I haven't investigated Linux/Gtk.

The timestamp of the event is related to other relevant timestamps in the system, including sound and audio. Synchronization is a *very* tricky problem. It's probably out of scope to try to fully solve it now, but we should keep it in mind so we don't paint ourselves into a corner. One highly relevant source of timestamps is frame presentation statistics, for example [DXGI_FRAME_STATISTICS](https://docs.microsoft.com/en-us/windows/win32/api/dxgi/ns-dxgi-dxgi_frame_statistics) on Windows. Ideally, these should also be used instead of the current time for generating animation intervals.

There are many open questions. Should we have both a "posted" and "retrieved" timestamp? Should we be trying to do sophisticated heuristics (something roughly like a clawback filter) to synchronize the various disparate sources of timestamp data? Discussion is welcome, and this might be a good issue for someone who loves a deep challenge.

Start with the issue's platform references for Windows MSG.time, macOS NSEvent.timestamp, and the uninvestigated Linux/Gtk path. Resolve the open questions about timestamp type, posted versus retrieved times, cross-platform conversion, and synchronization with related system timestamps. Done requires an agreed design rather than a single obvious edit.

Written by the indexing model from the issue text.