fix: do not error for remote chart if there's a missing local repo cache

[mikeseese]

What this PR does / why we need it:
This PR closes #11961 by continuing to scanReposForURL if one of the local repos doesn't have a cache rather than erroring. This PR also adds the hint to helm repo update to ErrNoOwnerRepo to catch cases that are expecting to find it in the local cache (i.e. u := localrepo/cache).

Ultimately, this makes it so that a full HTTP URL for the chart will never fail because in chart_downloader.go it will proceed to downloading it if scanReposForUrl returns ErrNoOwnerRepo (line 216):

If there is no special config, return the default HTTP client and swallow the error.

This PR also adds a test that tests for the case where there's a non-empty repository config but an empty repository cache.

Special notes for your reviewer:
I verified this actually fixes #11961 by using the reproduction steps listed there compared with the released/installed version of helm and the manually built version located in bin/helm (after running make build). Here was the final output of that:

/mnt/d/work/helm$ helm template traefik --repo https://helm.traefik.io/traefik
Error: no cached repo found. (try 'helm repo update'): open /home/seese/.cache/helm/repository/ingress-nginx-index.yaml: no such file or directory
/mnt/d/work/helm$ bin/helm template traefik --repo https://helm.traefik.io/traefik
---
# Source: traefik/templates/rbac/serviceaccount.yaml
kind: ServiceAccount
apiVersion: v1
metadata
<...>

If applicable:

• this PR contains documentation
• this PR contains unit tests
• this PR has been tested for backwards compatibility

[mikeseese]

I'm moving this back to draft to add in some minor changes to increase performance (as suggested by @MichaelMorrisEst in #11961).

[mikeseese]

Looking at the comments in chart_downloader.go, I think the original intention of scanReposForURL is to handle more than just finding the first repo in repositories.yaml that has the same URL. I don't have time to ensure the changes are not introducing a regression, so I'm leaving the PR as is. This ensures the least amount of regressions, is easier to grok/approve, requires less tests, solves my specific use cases, etc. in favor of leaving "making scanReposForURL" more performant for another, more determined contributor.

[MissedTheMark]

Change looks good, as does the new test.

Manually testing the change locally shows that it works correctly.
I have also verified that the new test fails on the old code, and passes with this change.

New test

Test failing on old code:

Test passing on new code:

Manually testing locally

Before this change:

After adding a repo (ingress-nginx) then clearing the Helm cache (note the unexpected reference to ingress-nginx-index.yaml):

$ helm template traefik --repo https://helm.traefik.io/traefik
Error: no cached repo found. (try 'helm repo update'): open /Users/mark/Library/Caches/helm/repository/ingress-nginx-index.yaml: no such file or directory

After this change:

After adding a repo (ingress-nginx) then clearing the Helm cache:

$ helm template traefik --repo https://helm.traefik.io/traefik
---
# Source: traefik/templates/rbac/serviceaccount.yaml
kind: ServiceAccount
apiVersion: v1
metadata:
  name: release-name-traefik
 ... etc

[MissedTheMark]

@mikeseese Looks like there's some CI issue, might be worth a merge from main?

[mikeseese]

Thanks @MissedTheMark for the review! I just rebased off of main. Let me know if there are any other issues.