cache last callstack for further callstack walking performance improvement

_From [bruen...@google.com](https://code.google.com/u/109494838902877177630/) on April 16, 2013 19:53:10_

Callstack walking for optimized code is complex and I could cite quite a few issue #s here on tweaks for both functionality and performance. This issue covers going a step beyond the work in issue #1186 and caching the full last callstack, with fp's, to try and avoid find_next_fp() even more. It still shows up on perlbench diffmail:

1833.59user 5.72system 30:45.28elapsed 99%CPU (0avgtext+0avgdata 429768maxresident)k
ITIMER distribution (182407):
0.0% of time in APPLICATION (1)
3.2% of time in INTERPRETER (5785)
0.3% of time in DISPATCH (628)
0.1% of time in SYSCALL HANDLER (103)
3.0% of time in INDIRECT BRANCH LOOKUP (5421)
43.8% of time in FRAGMENT CACHE (79891)
49.7% of time in UNKNOWN (90578)
RES-pcsamples.0.7276.html
897 get_shadow_table
1011 add_to_delay_list
1109 bitmapx2_set
1262 find_free_list_entry
1307 packed_callstack_hash
1438 safe_read
1662 module_lookup
2334 shadow_set_range
2458 address_to_frame
2999 rb_in_node
3925 print_callstack
5895 find_next_fp

_Original issue: http://code.google.com/p/drmemory/issues/detail?id=1187_