.NET Reactor: Fix CflowConstantsInliner corrupting exception handlers

[GarethWright]

Problem

CflowConstantsInliner.InlineAllConstants replaces whole Instruction objects when inlining the cflow constant fields:

instrs[i] = Instruction.CreateLdcI4(value);
if (nopNext)
    instrs[i + 1] = Instruction.Create(OpCodes.Nop);

If the original ldsfld/ldfld happens to be an exception-handler boundary (e.g. HandlerEnd) or a branch target, the handler/branch keeps referencing the old, now-detached instruction, which is no longer in body.Instructions.

InstructionListParser then throws KeyNotFoundException from instrToIndex[...] (via MarkAsBranchTarget / GetInstrIndex) when it rebuilds that method's blocks. Because this happens during normal method deobfuscation, it aborts deobfuscation of the entire assembly, not just the affected method.

Observed stack trace against current master:

ERROR: Stack trace:
ERROR:    at de4dot.blocks.InstructionListParser.MarkAsBranchTarget(Instruction instr) in InstructionListParser.cs:line 58
ERROR:    at de4dot.blocks.InstructionListParser.CreateExceptionBranches() in InstructionListParser.cs:line 68
ERROR:    at de4dot.blocks.InstructionListParser..ctor(...)
...
ERROR:   System.Collections.Generic.KeyNotFoundException

Fix

Write the constant in place (OpCode / Operand) instead of substituting new Instruction objects, so exception handlers and branches keep pointing at a live instruction. The emitted IL is unchanged — only object identity is preserved.

Trigger condition

The bug isn't specific to a particular .NET Reactor version. It only requires an inlined cflow-constant field access to coincide with an exception-handler boundary or a branch target, which is why it shows up on some protected assemblies and not others.

Testing

• On an affected assembly (a protected binary where an inlined constant field sits on an exception-handler boundary): KeyNotFoundException crash → deobfuscates and saves normally.
• No regression across a corpus of ~50 other .NET Reactor-protected assemblies: all still save, no new crashes.

[GarethWright]

Closing this and reopening as a fresh PR with a tidied-up description. No change to the code.