Set of interpreter related fixes for coreclr tests

[janvorli]

The changes in this PR were made to fix various issues uncovered by the coreclr tests.

• Handle case when a try block is eliminated because it can never be executed.
• Ensure call stubs are generated for IL stubs.
• Fix shadow stack updating during stack walk
• Fix handling managed exceptions thrown as SEH from managed code called by JIT / Interpreter compiler in interpreted code.
• Fix resuming after catch in interpreted code for cases when interpreted code calls another interpreted code via a prestub, which currently happens for delegates.
• Ensure that ip is saved to the pFrame->ip before calling native methods that can throw to make sure that the stack walk reports correct return IP.

[dotnet-policy-service]

Tagging subscribers to this area: @BrzVlad, @janvorli, @kg
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull Request Overview

This PR fixes various issues discovered by the coreclr tests related to interpreter functionality, focusing on exception handling, stack management, and IL stub generation. The changes primarily address edge cases in exception propagation and ensure proper state management during method calls.

Key changes:

• Adds proper handling for managed exceptions thrown from PrepareInitialCode calls
• Fixes shadow stack pointer (SSP) updating during stack walks in interpreter frames
• Ensures call stubs are generated for IL stubs when using the interpreter

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
src/coreclr/vm/interpexec.cpp	Adds exception handling wrapper, fixes IP saving timing, and improves catch resumption logic
src/coreclr/vm/frames.cpp	Updates shadow stack pointer during register display updates
src/coreclr/vm/dllimport.cpp	Ensures call stub creation for interpreted IL stubs
src/coreclr/interpreter/compiler.cpp	Handles unreachable try blocks and validates leave instruction targets

[davidwrighton]

LGTM. I'm convinced we have bugs around GC reporting here due to moving the ip early, but that is going to require a trickier fix than I had originally thought. My current thought is that for GC reporting purposes we should do a tweak where we report the conservative merge between the stack before/after the current IP when at the top of a series of interpreter stack frames, but if the InterpMethodContextFrame currently being reported isn't the active frame on the InterpreterFrame we can report only the current data.