Optional per-node dedup identity for sharing one database across otherwise independent nodes that run the same actors.

This is orthogonal to Cyclium.StackSlug:

• stack slug selects which actors run (host-app placement) and scopes recovery — different stacks already run differently-named actors, so their dedup keys differ via actor_id.
• env distinguishes otherwise-identical nodes so two :full nodes running the same actors against the same DB (e.g. a developer's laptop and a hosted test deployment, or an RC node sharing prod's DB) operate as independent worlds: each creates and runs its own episodes, recovers only its own orphans, and keeps its own findings.

When env is set it scopes work along three dimensions; when unset, behavior is byte-identical to single-env operation (legacy NULL rows belong to the unset/default env):

1. Dedup/claim keys — folded into the episode dedupe_key (which the work-claim lease and derived output keys inherit) and explicit output keys via scope_key/1.
2. Recovery — episodes and workflow instances carry source_env (mirroring source_stack); Cyclium.Recovery only sweeps/reconciles its own env, matched by strict equality so an env-tagged node never resurrects the default node's work.
3. Findings — each env keeps its own active finding per key (the cyclium_findings.env column + a [finding_key, status, env] unique index); Cyclium.Findings reads and upserts are cordoned to the current env.

An RC node configured as :full with recovery on, tagged env: "rc" and sharing prod's DB, can therefore drive UI interactions and run actors fully independently of the prod node (env unset) without either deduping, recovering, or clearing the other's work.

Configuration

# Set per node (e.g. from a CYCLIUM_ENV env var in runtime.exs). Leave
# unset on the shared/hosted node; set it on the isolated (dev) node.
config :cyclium, :env, "dev-jane"