The variable is `Version`, not `Branch`, and if we're using the template
variable syntax, it should have the leading dot that it needs to.

Noticed we didn't include these fields in our prometheus and logging
integration.

Test Plan: go test ./...

When the user specifies an invalid mirror config that contains an entry
without any of the GithubUser / GithubOrg / ... repository definitions,
zoekt-indexserver would crash, because `cmd` stays `nil` and causes a
SIGSEGV when it is later accessed in `main.executeMirror`.

Test Plan: go test ./...

Useful in local testing to capture both on and off cpu time spent.
Should consider shipping this in the webserver as well.

Test Plan: ran zoekt with -full_profile flag and inspected output in
pprof.

We have been running btree as the default for many months. We worried
about a performance hit, but it never happened. After some recent local
testing I did I noticed the btree actually interacted with the disk more
efficiently. So the old code both uses more memory and is slower, lets
just remove it.

Test Plan: go test ./...

We believe this will improve performance of the btree lookups. We are
investigating this to make it faster to rule out a shard (when freq==0).
Testing locally on a large corpus we halved the time spent in IO.

Locally Sort shows up in the profiles significantly, but there are two
facts mitigating that:
- Locally my file page cache is primed so IO rarely is going to disk.
- We likely will implement an IR for Zoekt which will amortize the Sort
  to once per search rather than once per shard.

Test Plan: go test ./... and performance profiling via via ./cmd/zoekt.

Co-authored-by: Stefan Hengl <stefan@sourcegraph.com>