fix: prevent excessive repo-server disk usage for large repos (#8845)

[crenshaw-dev]

Fixes #8845

Fetching specific commits allowed us to get commits pointing to non-standard refspecs (like refs/pulls/*/head).

But it costs a lot more disk space (demonstrated in the issue) and slows down app refreshes after commits.

This change defaults to the old behavior (fetch standard refspecs, checkout commit) and then falls back to fetching a specific commit if the checkout fails.

So if someone has an Application pointing to HEAD of a repo with 100k commits, this will efficiently fetch each commit without filling up the disk.

And if someone has an application pointing to refs/pulls/123/head, this will fail on the initial fetch and then fall back to fetching the specific commit. It's higher latency, but I think it's justified to avoid filling up the disk.

I've also confirmed there is no regression by testing this app:

apiVersion: argoproj.io/v1alpha1
kind: Application
metadata:
  name: argo-cd
  namespace: argocd
spec:
  destination:
    name: ""
    namespace: default
    server: https://kubernetes.default.svc
  project: default
  source:
    path: manifests/core-install
    repoURL: https://github.com/argoproj/argo-cd.git
    targetRevision: refs/pull/8897/head

[codecov]

Codecov Report

Merging #8897 (a595331) into master (b1ff9db) will increase coverage by 0.03%.
The diff coverage is 69.23%.

@@            Coverage Diff             @@
##           master    #8897      +/-   ##
==========================================
+ Coverage   43.10%   43.13%   +0.03%     
==========================================
  Files         186      186              
  Lines       23359    23359              
==========================================
+ Hits        10069    10077       +8     
+ Misses      11849    11841       -8     
  Partials     1441     1441              

Impacted Files	Coverage Δ
reposerver/repository/repository.go	60.08% <69.23%> (+0.85%)	⬆️
util/settings/settings.go	48.10% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b1ff9db...a595331. Read the comment docs.

[crenshaw-dev]

Note: planning to run this internally for a week to see if it solves an internal problem we faced. This is sensitive code with a lot of edge cases, so we should be careful.

[alexmt]

Thanks a lot @crenshaw-dev ! Internal testing was successful. LGTM