fix(resume): --from-step with --detach creates 3 phantom run records

[nextlevelshit]

Bug

Running wave run --detach --from-step <step> <pipeline> creates 3 separate pipeline_run records instead of 1, and the run detail pages for the parent runs show empty steps/events.

Root Cause

Two compounding bugs:

Bug 1 — cmd/wave/commands/run.go

The --detach path creates a run record, then spawns a subprocess with --run <id> --from-step <step>. The subprocess hits this condition:

```go
if opts.RunID != "" && opts.FromStep == "" { // false when --from-step is also set
runID = opts.RunID // skipped
} else if store != nil {
runID, err = store.CreateRun(...) // creates a 2nd record
}
```

Because --from-step is set, the pre-created --run ID is ignored and a second record is created.

Bug 2 — internal/pipeline/resume.go

ResumeFromStep() always calls createRunID() which calls store.CreateRun(), creating a 3rd record, even though the executor already has e.runID set via WithRunID.

Result

• 3 runs in the dashboard for one logical operation
• The 2 parent runs show 0 steps, empty events, wrong status
• Actual work happens in the 3rd (leaf) run under a different ID than what the user sees

Fix

• Always reuse opts.RunID when it is set (remove the && opts.FromStep == "" guard)
• In ResumeFromStep, reuse r.executor.runID when non-empty instead of calling createRunID()