commit: remove parse_commit_no_graph()

[derrickstolee]

The parse_commit_no_graph() method was added in 43d3561 ("commit-graph
write: don't die if the existing graph is corrupt" 2019-03-25) as a way
to avoid persisting bad data across commit-graph files. That is, if the
commit-graph file has undetected corrupt data -- such as a flipped bit
in a parent int-id value -- then that data will persist to the next
commit-graph file. The parse_commit_no_graph() method was used to always
use the pack data directly instead.

Unfortunately, this comes at a significant performance cost. In both
time and memory, parsing from pack files is much slower than parsing
from the commit-graph file. In a repository with 4.5 million commits,
this can lead to Git taking up to 11gb of memory to rewrite the file.

Now that the incremental commit-graph file format exists, we can rely
on the quality of the commit-graph file if we follow the two-step
pattern of (1) write a commit-graph with "--split" and (2) run "git
commit-graph verify --shallow" to verify the tip file.

---

@jrbriggs, @jeffhostetler, @jamill and others: this change should revert
the performance problems we are seeing with the M153 release. I will
test this carefully after generating a Windows installer. I'm not sure if
this change would be something we can send upstream or not, but I
will start a conversation about it on the list.

[derrickstolee]

(I notice of course that this will fail the test that he added for this case... I will probably disable it for now.)

Edit: I just did a revert (and adjusted the conflicts around changes on top of that commit).