Introduce 'git backfill' to get missing blobs in a partial clone (#5172)

This change introduces the `git backfill` command which uses the path
walk API to download missing blobs in a blobless partial clone.

By downloading blobs that correspond to the same file path at the same
time, we hope to maximize the potential benefits of delta compression
against multiple versions.

These downloads occur in a configurable batch size, presenting a
mechanism to perform "resumable" clones: `git clone --filter=blob:none`
gets the commits and trees, then `git backfill` will download all
missing blobs. If `git backfill` is interrupted partway through, it can
be restarted and will redownload only the missing objects.

When combining blobless partial clones with sparse-checkout, `git
backfill` will assume its `--sparse` option and download only the blobs
within the sparse-checkout. Users may want to do this as the repo size
will still be smaller than the full repo size, but commands like `git
blame` or `git log -L` will not suffer from many one-by-one blob
downloads.

Future directions should consider adding a pathspec or file prefix to
further focus which paths are being downloaded in a batch.

Author: dscho

Documentation/git-backfill.adoc

@@ -9,7 +9,7 @@ git-backfill - Download missing objects in a partial clone
 SYNOPSIS
 --------
 [verse]
-'git backfill' [--batch-size=<n>] [--[no-]sparse]
+(EXPERIMENTAL) 'git backfill' [--batch-size=<n>] [--[no-]sparse]
 
 DESCRIPTION
 -----------

builtin/backfill.c

@@ -26,7 +26,7 @@
 #include "path-walk.h"
 
 static const char * const builtin_backfill_usage[] = {
-	N_("git backfill [--batch-size=<n>] [--[no-]sparse]"),
+	N_("(EXPERIMENTAL) git backfill [--batch-size=<n>] [--[no-]sparse]"),
 	NULL
 };