PE: Introduce a fast path for unwinding stack only

[ishitatsuyuki]

#13 introduced support to recover full register information from unwind info, not only RIP, RSP and RBP. However, this does have some performance implications:

• "Pop register" opcodes still reads the stack to recover general purpose registers even though they will not be used.
• "Save register" opcodes can be skipped and take the unwind rule fast path if register recovery is not needed. On the other hand the offsets are likely too large to encode into an unwind rule if register recovery is needed.

There should be some way to select either the faster behavior or the register recovery behavior.