Compiles advice injectors down to Emit

[plafer]

Closes #1457

Left to do:

• Rename AdviceInjector to NativeEvent (or NativeAdviceEvent?)

[plafer]

Last thing we could do in this PR is rename AdviceInjector to NativeEvent or something like that

[bobbinth]

Looks good! Thank you! I left some comments inline. Most are pretty small and the bigger once can be done in future PRs.

Once this and #1572 are merged (I would probably first merge this into #1572 and then merge #1572), we can close #1457. But let's also create an issue describing the remaining work on this (e.g., getting rid of AdviceInjector enum, moving some events to the standard library, coming up with a mechanism for handling events in an extensible manner etc.).

[bobbinth]

Not for this PR, but in the future we should probably split event_id into source and ID where source identifies a library (e.g., source=0 for the system events, source=1 for the standard library etc.). We've discussed some of this in #1457, but may be good to create a separate issue once #1457 is closed.

[plafer]

My issue with this though is that it reduces the number of "non-colliding libraries" from potentially 2^32 down to 2^16. Also, most libraries will probably only define a few events, wasting ~2^16 bits of unused event IDs (but still allocated to them).

I figured it to be better to just have libraries subscribe to the set of event IDs they need, (which the host can track with a BTreeMap<u32, EventHandler>). Then, we're actually very unlikely for libraries to define the same event ID (assuming all libraries properly generate their event IDs randomly).

[bobbinth]

Let's discuss this in a follow-up issue. For example, we could still have 32-bit space for libraries and then within each library we could have a 16-bit space for event IDs.

[bobbinth]

Probably not for this PR, but I'm wonder if we actually need the AdviceInjector enum at all. I think pretty much in all cases it can be replaced with Operation::Emit(EVENT_ID) and the code would be just as readable.

[plafer]

I renamed AdviceInjector to SystemEvent, and I don't think it's so bad to have an enum. For example, in handle_system_event, we match on an enum instead of u32 constants, so if we add a SystemEvent, we'll get a compile error there (instead of nothing)