Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

These should now let us reconstruct the commit object (or at least the
git cat-file equivalent) appview-side.

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: oppiliappan <me@oppi.li>

this also populates the script with the correct executable path for
the `knot` command.

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: oppiliappan <me@oppi.li>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>

Signed-off-by: oppiliappan <me@oppi.li>

performing rev-list and filetree in sequential occupy a large amount of
the client budget. we can parallelize all these expensive operations.

Signed-off-by: oppiliappan <me@oppi.li>

uses `git log --name-only` to speed things along. the performance
tradeoff here is that active repositories tend to load last-commit times
really quickly, but directories with one file that was modified 20k
commits ago will result in the entire log being walked through.

git-log command is parsed as it streams to avoid having to wait for the
entire command to complete, and as soon as the entire directory's
listing is hydrated, we cancel the command. most of the time, this
results in hydration of subdirectories also (this gets cached),
resulting in good experience when browsing a repo (typically moving down
directories gets faster with each click, as fewer files need hydration).

Signed-off-by: oppiliappan <me@oppi.li>

RepoIndex and RepoLog should no longer take longer on bigger
repositories. all endpoints are still backwards compatible.

Signed-off-by: oppiliappan <me@oppi.li>