Change InferType() to an abstract method and implement for every ILInstruction.
With this change, we now always have enough information to create a variable of an appropriate type to store the result of evaluating the instruction.
This previously was not the case for instructions producing "other value type", for which the stacktype-based fallback incorrectly produced `object`.
* Added additional code to remove the conv instruction present in the initialization part of the pinned region.
* Extended the code responsible for removing the unpin stloc to correctly match the inverted condition found in MCS 2.6.4 compiled code.
* Enabled already present correctness test to run for MCS 2.6.4.
This is a more generalized version of the fix on PR #3110 proposed by @ElektroKill.
These blocks could trigger assertions if LoopDetection was creating a loop BlockContainer from them (BlockContainers have an assertion requiring an ILRange).
Closes#2533 and #2457.
Let's distinguish between the original pinned locals and the PinnedRegion locals.
The format need declarations if any are left over after transformations; the latter don't.
This removes an error condition where pinned variables could remain after DetectPinnedRegions, resulting in invalid C#.
The downside is that it's possible for the code size to increase exponentially when given sufficiently evil IL code.
Not every pinned region has a clean `P = null` assignment to mark its end.
If a second pinned region starts with the same variable `P`, consider that to mark the end of the previous pinned region for that variable.
Also, fix a bunch of special cases with empty pinned regions.
This also makes CleanUpTryFinallyAroundPinnedRegion() redundant as it is no longer necessary to trigger the array-to-pointer transform; so the normal elimination of pinned variable resets is sufficient.