#2520 Versions written by different steps of one workflow run all stay isLatest=true (data query / readModelData / queryData return stale "latest")

Swamp 20260925.013241.0-sha.4489c40d (also 20260922.185723.0). Same symptom as #1902 (shipped), via a path that fix doesn't cover.

Repro (fresh repo)

Model @repro/counter with one resource spec item and a method write that does ctx.writeResource("item", "item-" + args.name, { state: args.state, step: args.step }).

Workflow two-steps: one job, step s1 runs c1.write {name: b, state: Ingesting}, step s2 (dependsOn s1) runs c1.write {name: b, state: Ingested}.

swamp workflow run two-steps
swamp data query 'modelName == "c1" && name == "item-b"' --select '"v=" + string(version) + " latest=" + string(isLatest)'

Actual: v=1 latest=true and v=2 latest=true (distinct ids and content; they differ only in tags.step/stepName). Expected: only v2 is latest.

Scope

• Only writes from different steps of the same workflow run. 3 steps across 2 jobs → 3 "latest" versions.
• Fine: two swamp model method run calls; two separate runs of a 1-step workflow; CLI → workflow; workflow → CLI.
• Wrong: swamp data query, extension context.readModelData(model, spec), context.queryData(...). A later step in the same run already sees the duplicates.
• Right: swamp data get, swamp data list, swamp data versions (reports v1 isLatest=false), context.readResource, CEL data.latest(...).
• swamp data gc doesn't clear it. The next write from outside that run demotes the old versions, but that run's own multi-step writes are again all left latest.

Impact

An extension that lists its resources with readModelData and filters on content (e.g. state == "Ingesting") matches the stale version. If it then writes both records back, it fails with Data output validation failed: Duplicate data instance name. We work around this by keeping the highest version per name.