No commit message

No commit message

No commit message

When a dollop with small fallthroughs would fit entirely in an 
allocated region but some of those fallthroughs are already placed,
coalescing would miscalculate what to do. It would say, "Let's
place it all" but would eventually want to put a jmp to the 
already-placed dollop. When the fallthoughs were small, that 
jump would be bigger than the entirety of dollop and cause
us to overflow the reserved space.

Add plugin support for
1. Plugins to retarget pins

2. Plugins to disallow coalescing dollops that they address.

1. Allow AddressDollop() to tell Zipr whether
the dollops should be coalesced. Defaults to true.

2. Allow the plugin to retarget a pin.

1. A plugin that returns space smaller than "min size"
is not bad if the dollop fits within the space entirely.

2. GetNearbyFreeRanges() should really treat the parameter
as a lower bound.

Previous code calculated the dollop based on
the comprising instructions' fallthrough. This
is incorrect in the case where a dollop stops
short (because it was split during placement,
ran through an IBT, etc).

This is important for a number of reasons. 
Most importantly, however, is the fact that it
will cause a problem with the push64 relocations
calculations.

Order the dollops in the placement q according to
their original address in the program.

Give the source address for the instruction
that is targeting the dollop to be placed. This
will give another signal to placement plugins
to help them make a decision.