fix(ci): Stop running multiple full syncs on different branches

[teor2345]

Motivation

We are accidentally running about 8 simultaneous full syncs on testnet, which is slowing down testnet.

Specifications

https://docs.github.com/en/actions/using-jobs/using-concurrency

https://docs.github.com/en/actions/learn-github-actions/contexts#needs-context

Solution

Fix long job concurrency:

• Only run one checkpoint/full/lwd sync at a time, by removing the branch from the concurrency group label
• Allow a manual sync to run at the same time as automatic syncs, by adding the manual sync setting to the label

Fix cache-using job conditions:

• If cache generation jobs are skipped due to concurrency, and the cache is missing, also skip their dependent jobs, rather than failing them when the disk isn't found
• This is implemented by requiring that the disk was found in the cache, or the disk was previously generated in this workflow run by a successful job

Related fixes:

• Remove outdated concurrency and fix docs on the lightwalletd send transactions test, which now runs on every PR

Review

This is urgent because we're putting a lot of load on testnet.

See my comment about manual testing of this fix below.

Reviewer Checklist

• Will the PR name make sense to developers?
• Are the PR labels correct?
• Does the code do what the ticket and PR says?

Follow Up Work

I'm going to cancel all the extra jobs now, and just leave the one in this PR, the release PR #6632, and the one that's about to finish.

[teor2345]

I did a manual test of these changes. In this workflow, the full testnet sync was skipped due to concurrency, and so the testnet checkpoint generation was also skipped:
https://github.com/ZcashFoundation/zebra/actions/runs/4944180938/jobs/8839453807

This is the correct behaviour. (Previously the checkpoint generation job would fail.)

[gustavovalverde]

I agree with this as a bugfix, but I have a "feeling" that we should move the testnet sync to the main branch, and just trigger it on PRs on-demand (using a label or comment).

[teor2345]

I agree with this as a bugfix, but I have a "feeling" that we should move the testnet sync to the main branch,

it is already just running on the main branch

and just trigger it on PRs on-demand (using a label or comment).

it only runs on PRs when there is no cached state disk, or when manually triggered using workflow dispatch.