Add misc. optimizations to filter tool

This CL adds 3 new peephole optimizations:

save/clipR/drawR/restore where the rect is inside the clip
save/clipRR/drawR/restore where the clip is inside the rect
save/clipR/drawBitmap/restore where the clip == the dest rect

Here are the firing rates for the new peephole optimizations:

opt 2: 441 save/clipR/drawR/restore
opt 3: 222 save/clipRR/drawR/restore
opt 4: 769 save/clipR/drawBitmap/restore

Executing desk_amazon.skp                2: 4        4: 6
Executing desk_blogger.skp                     3: 1  4: 29
Executing desk_booking.skp                     3: 2  4: 14
Executing desk_br337.skp                             4: 35
Executing desk_ebay.skp                  2: 4        4: 21
Executing desk_espn.skp                  2: 28       4: 1
Executing desk_facebook.skp              2: 25 3: 3  4: 78
Executing desk_gmailthread.skp           2: 2  3: 4
Executing desk_googleplus.skp            2: 39 3: 4  4: 6
Executing desk_googlespreadsheet.skp     2: 8  3: 1
Executing desk_googlespreadsheetdashed   2: 8  3: 1
Executing desk_gws.skp                   2: 1  3: 4  4: 10
Executing desk_linkedin.skp                    3: 4
Executing desk_pinterest.skp             2: 4  3: 2  4: 49
Executing desk_sfgate.skp                2: 5  3: 2  4: 16
Executing desk_techcrunch.skp            2: 1        4: 5
Executing desk_twitter.skp               2: 6  3: 1  4: 3
Executing desk_weather.skp               2: 9        4: 7
Executing desk_wordpress.skp                   3: 9  4: 9
Executing desk_wowwiki.skp               2: 6  3: 12 4: 6
Executing desk_yahooanswers.skp          2: 27 3: 1  4: 21
Executing desk_yahoogames.skp            2: 6  3: 14 4: 5
Executing desk_yahoonews.skp             2: 12 3: 15 4: 23
Executing desk_yahoosports.skp           2: 19 3: 10 4: 13
Executing desk_youtube.skp               2: 70 3: 10 4: 18
Executing mobi_wikipedia.skp                         4: 8
Executing tabl_androidpolice.skp               3: 6
Executing tabl_cnet.skp                  2: 2        4: 7
Executing tabl_cnn.skp                   2: 2  3: 4  4: 54
Executing tabl_cuteoverload.skp          2: 3  3: 50 4: 4
Executing tabl_deviantart.skp                  3: 4
Executing tabl_digg.skp                  2: 22       4: 19
Executing tabl_engadget.skp              2: 27 3: 24 4: 27
Executing tabl_frantzen.skp              2: 2        4: 3
Executing tabl_gamedeksiam.skp           2: 6        4: 108
Executing tabl_googleblog.skp            2: 1  3: 24 4: 19
Executing tabl_googlecalendar.skp        2: 1  3: 1  4: 5
Executing tabl_gspro.skp                 2: 1  3: 1  4: 2
Executing tabl_hsfi.skp                  2: 16 3: 1  4: 39
Executing tabl_mlb.skp                   2: 1  3: 2  4: 3
Executing tabl_mozilla.skp                           4: 8
Executing tabl_nofolo.skp                            4: 1
Executing tabl_nytimes.skp                           4: 3
Executing tabl_onlinewsj.skp             2: 2  3: 1
Executing tabl_pravda.skp                2: 6        4: 10
Executing tabl_sahadan.skp               2: 3        4: 4
Executing tabl_slashdot.skp                          4: 4
Executing tabl_techmeme.skp              2: 3
Executing tabl_theverge.skp              2: 48 3: 2  4: 11
Executing tabl_transformice.skp                      4: 7
Executing tabl_ukwsj.skp                 2: 5  3: 1  4: 2
Executing tabl_vnexpress.skp             2: 3        4: 34
Executing tabl_worldjournal.skp          2: 3  3: 1  4: 12

[DerekS, 2013-03-11 18:09:11]

https://codereview.appspot.com/7705043/diff/2001/debugger/SkDebugCanvas.h
File debugger/SkDebugCanvas.h (right):

https://codereview.appspot.com/7705043/diff/2001/debugger/SkDebugCanvas.h#new...
debugger/SkDebugCanvas.h:98: SkTDArray<SkDrawCommand*>& getDrawCommands();
do you still need the const method if you have this one? 

Would a better solution be to have a setDrawCommands method that would replace
the array? That way if we do cache anything inside the DebugCanvas we know that
it needs to be invalidated?

[robertphillips, 2013-03-11 18:46:43]

I left the const version since no one but the filter tool requires write access.

The current structure of the DebugCanvas hampers the iterative application of
optimizations. I think where we have to go is to have an interface like:
 DeleteCommand(int command);
 InsertCommand(int index, Command*);
and hide the command array entirely.

[DerekS, 2013-03-11 19:59:22]

lgtm if you comment you add a comment to the new SkDebugCanvas header that this
function is not intended for future use.  Going forward when we have another
potential client we can refactor to meet the needs of both.

[robertphillips, 2013-03-11 22:53:43]

Done.

Committed as r8087.