Make ctx.pass immutable

[Simn]

This was easier than expected. We now also clone the context before typing expressions, which means that we don't have to mutate ctx.pass <- PTypeField anymore. This in turn means that all context instances now have a fixed pass, which should make everything much easier to debug. We could even give them names and go to their piano rehearsals.

I'd like to see if I can break the macro change here because I feel like this should be breakable, given that we no longer enter PTypeField pass. I think this portion of code should also clone a new expression context, but before I do that I'll see how to actually trigger an issue here.

[Simn]

This indeed fails exactly like I expected it to: a build macro causing the typing of a module, and then trying to access a class field doesn't work because we never flush PBuildClass. It's a bit scary that we didn't test this before, but at least the typer passes behave predicable for once.

[kLabz]

Wartales breaks badly with this change :/ (will send errors via slack tomorrow morning)

Other projects seem mostly ok (so it may be about hxbit closures?), ie. as ok as they are with current nightlies (so need to revert @:bypassAccessor and bind changes in most projects, and there are some old (pre-hxb at least) issues with complex recordings; but results are the same as with dev for replays and compilation is ok).

[Simn]

For documentation purposes: the problem was that ctx.allow_inline and ctx.allow_transform weren't inherited by "child" contexts. The expected state appears to be that they're reset only when creating a new module's context.