Fix several problems related to perf_debug_verbose.

[wujingyue]

1. The name group_id is overloaded. Sometimes it refers to the index in run order, sometimes SegmentedGroup::groupId(). That indeed confused this code, partially causing the bug. I cleaned up the naming so it's less likely to make such bugs.
2. The number of output bytes processed is only written under execute_kernel_, but is read (e.g.) even when the kernel is not executed. So I hoisted outputBytesProcessed out of if execute_kernel_.

Fixes #1943.

[wujingyue]

!build

[wujingyue]

!build

[jacobhinkle]

This change means that we print bytes processed even when we do not execute the kernel, correct?

[wujingyue]

Correct -- I do see gb_per_s=inf printed, which is the right answer.

I believe this will change again with #1930. This printing will have to be skipped because kernel_id_ won't even exist.

[jacobhinkle]

LGTM

[wujingyue]

!build

[wujingyue]

!build

[wujingyue]

!build