[kyoresuas]

Closes #886.

BACKUP_JITTER was applied unconditionally in runScript, so a one-off run like docker exec <container> backup was delayed by a random amount of time, too. As discussed in the issue, jitter now only applies to runs that are triggered by a cron schedule.

• runScript takes a new runScriptOpts argument (same pattern as foregroundOpts) with a scheduled flag.
• The cron callback in command.schedule passes scheduled: true, runAsCommand (one-off run) leaves it unset.
• The delay calculation moved into a small jitterDelay helper so it can be unit tested without a Docker daemon or a file lock. The behavior for scheduled runs is unchanged.
• The reference docs now say that jitter only applies to scheduled runs. I also replaced the dangling comma at the end of the first sentence of that comment block with a period.

Setting BACKUP_JITTER=0m on invocation as a workaround is no longer necessary, but still works.

Tested:

• go test ./cmd/... ./internal/..., go build ./... and go vet ./cmd/... ./internal/... pass, gofmt -l reports nothing.
• Added TestJitterDelay covering a manual run with jitter configured (no delay), a scheduled run with zero and negative jitter (no delay), and a scheduled run with jitter configured (delay always within [0, max]).
• I did not run the integration tests in test/ as they need Docker.

[m90]

This looks good, thank you! There's a couple of things in flight right now, so I'll merge this whenever all of this is ready for release.