Make SkDeque::back faster & inline

Maybe a tiny bit faster on our benchmarks:

        aaclip_build_path_AA 8888  7.41       7.63      -0.22    -3.0%
               dash_1_noclip 8888  9.44       9.70      -0.26    -2.8%
               dash_4_noclip 8888  3.55       3.52      +0.03    +0.8%
            aaclip_setregion 8888  7.19       7.07      +0.12    +1.7%
            nested_aaclip_AA 8888  3.18       3.01      +0.17    +5.3%
              aaclip_path_AA 8888  3.92       3.71      +0.21    +5.4%
              aaclip_rect_AA 8888  2.05       1.93      +0.12    +5.9%
               dash_0_noclip 8888  1.69       1.59      +0.10    +5.9%
            nested_aaclip_BW 8888  2.00       1.88      +0.12    +6.0%
        aaclip_build_path_BW 8888  3.85       3.58      +0.27    +7.0%
              aaclip_path_BW 8888  2.26       2.08      +0.18    +8.0%
        aaclip_build_rect_AA 8888  1.11       0.99      +0.12   +10.8%
        aaclip_build_rect_BW 8888  0.79       0.70      +0.09   +11.4%
              aaclip_rect_BW 8888  0.73       0.64      +0.09   +12.3%
              dash_1_clipped 8888  0.31       0.22      +0.09   +29.0%

[reed1, 2012-08-17 13:48:42]

fast, but slightly deep change. Do we feel secure with our current unittests?

lgtm

[robertphillips, 2012-08-17 13:56:01]

I feel secure. I actually implemented this in two phases where in the first I
left in the original front & back methods & just added asserts to validate that
the new functions worked. I ran this through all the GMs and tests and the new
entry points worked correctly. So ... maybe not as dramatic a switch as the
final code appears.

[robertphillips, 2012-08-17 13:56:25]

committed as r5149