Fix exception propagation over HW exception frame on macOS arm64

[janvorli]

There is a bug in setting up the fake stack frame for
the PAL_DispatchExceptionWrapper. The FP and SP were not saved
to the stack frame and the FP of the context was not set to
match the fake prologue. That caused failure to unwind over the
PAL_DispatchExceptionWrapper, reaching an unrelated native
function.

This change fixes it.

Close #62058

[janvorli]

@jkotas I've added a regression test for the issue (the mini repro from the #62058)

[janvorli]

I can see that two of the coreclr tests fail in the ci for this PR on osx arm64. I am investigating if that's related to this change.

[janvorli]

The fix is actually wrong. We were already storing the fp / lr on the stack a bit above the place I've added. The only thing was that instead of LR, we were putting there PC. Which was actually correct for everything except leaf functions (the issue #62058 has the exception in a leaf function, the VSD stub). Well, even for non-leaf functions, there would be a problem if the LR was used as a general purpose register between calls.
The real problem is that with the arm64 mechanism of making calls we cannot make unwinding of injected frame to get LR after the unwind to be different from the PC. The actual fix will need to do the same what we do for Linux - to make PAL_VirtualUnwind to recognize the PAL_DispatchExceptionWrapper and to extract the whole context at the failure location from the place where HijackFaultingThread has stored it.

[janvorli]

/backport to release/6.0

[github-actions]

Started backporting to release/6.0: https://github.com/dotnet/runtime/actions/runs/1669998117

[danmoseley]

@janvorli (you probably verified this already but) it does seem this broke the test, Should we open a bug?

WorkItems 
| where Started > now(-100d)  
| where FriendlyName == "System.Composition.Hosting.Tests"
| where ExitCode != 0
| where Status == "Timeout"
| join kind= inner (
   Jobs  | where Started > now(-30d) | project  QueueName , JobId, Build, Type, Source 
) on JobId
| summarize count() by ExitCode, QueueName, format_datetime(bin(Started, 1d), 'MM-dd')
| order by Started desc

ExitCode	QueueName	Started	count_
-3	osx.1015.amd64.open	01-10	4
-3	osx.1014.amd64.open	01-10	3
-3	osx.1015.amd64.open	01-09	13
-3	osx.1014.amd64.open	01-09	19
-3	osx.1015.amd64.open	01-08	9
-3	osx.1014.amd64.open	01-08	15
-3	osx.1100.arm64.open	01-08	2
-3	osx.1013.amd64.open	01-08	1

[krwq]

@danmoseley I think there is already an issue: #63561

[jkotas]

This is affecting too many PRs. We need to revert it: #63576