addronly: finer granularity or prefix shadow scheme

_From [bruen...@google.com](https://code.google.com/u/109494838902877177630/) on October 27, 2011 14:20:29_

original issue #394 prototype used simple 4-to-1 shadow mapping

*\* TODO 4B-to-1B will miss overflow in char\* and other non-size-aligned mallocs
should document it
makes it hard to test: since will miss the unaddr in malloc.c that tests
reading the padding, but will find the one in free.c.
need annotations!
or have separate .addronly.out template

***\* TODO how fix? worth fixing?

could use asan's prefix encoding
if do that, could back out to 8B-to-1B and use it for stack: though if only examine prefix on slowpath that's not a good idea.

could go back to 1B-to-2b.
change bit encoding to eliminate undef vs def for simplicity.
then need sub-dword code that gets the offs: no longer simple instru.
but simpler than full drmem since no prop: just need table lookup I guess.

these aren't the most serious of bugs: there's no other data there, just
padding, and technically app could call malloc_usable_size() and
legitimately use the data.

_Original issue: http://code.google.com/p/drmemory/issues/detail?id=650_