the rev-list approach to building format-patch internally performs
merge-base calculations.

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

the previous commit iterator approach was faulty in cases of merge
commits, as it only uses the first parent returned by .Parents().

this also makes use of the --no-merges flag when preparing commits for
format-patch.

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

when calling rev-list, HEAD was passed as a default argument:

git rev-list HEAD --count

but HEAD is only populated when we perform git.Open; it is zeroed out
when we perform git.PlainOpen:

git rev-list 00000... --count

this causes rev-list to exit with an error, when used in repos that are
git.PlainOpen'd.

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

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

similar to jetstream consumer, we now ingest events from every known
knot.

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

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

tracks commit per user by day

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

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

consumers can configure a cursor-store, where cursors of individual
event sources are stored. the module provides an in-memory store and a
redis-backed store.

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

- EventSource: a knot's identity (its hostname currently)
- Message: message struct that follows the event structure to
deserialize just rkey and nsid upfront, this could allow consumers to
filter by nsid or configure cursor ranges to listen to

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

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

initial commit does not have a parent commit and should be excluded from
the signature check.

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

this was incorrectly adding the current pull into the formatpatch twice.

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

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