Emit lane: empty then_/else_ arms emit bash no-op ':' instead of invalid syntax

[briansrls]

Summary

Follow-up to #7277. Pipeline.steps (v2.std.orchestration) is a plain FreeMonoid<PipelineStep> with no NonEmpty variant and no smart constructor, so a zero-step Pipeline (steps: Empty) is legally constructible today. Nested as an If's then_/else_ arm, orch_emit_if_step threaded that arm's emitted "" straight into the inline bash row's body slot, producing "if COND; then ; fi" — invalid bash (a then clause needs ≥1 command before the closing separator; real bash raises "syntax error near unexpected token `fi'"). This is a fabricated-invalid-output shape, not a refusal.

Zero production occurrences of an empty arm exist in the tree today (verified via Pipeline { steps: ... } census across dag/gunbc and src/v2/workflow) — this is a priced, cheap fix ahead of any real trigger, not a live incident.

Fix (uniform, no special-casing): an empty arm's body always emits the bash no-op : (orch_if_arm_body_or_noop in 05_emit_orchestration.dag), applied identically in the inline and block dispatch and regardless of whether the other arm is also empty. This covers four shapes with one rule:

• empty then_, else_ Absent
• empty then_, non-empty else_ present ("act only on the else")
• non-empty then_, empty else_ present ("act only on the then")
• both arms empty

Why no refusal for the both-empty case (reconsidered mid-review): Predicate (ExitZero/LogMatches/StrEq/etc.) executes a command — so even a both-empty If runs the condition and discards the outcome, a real "run cmd, ignore result" intent, not dead code. Refusing it would forbid working intent. This also matches the existing convention already in the codebase: a top-level empty Pipeline (not nested in an If) already emits "" cleanly via orch_emit_pipeline's HeadAbsent arm, with no error — so "empty pipeline is a legitimate no-op" was already this codebase's convention before this PR; the fix makes that convention hold uniformly whether or not the Pipeline is nested inside an If.

Option (3) declined, named on the carrier: a NonEmpty<PipelineStep> carrier on Pipeline.steps was considered (construction-first, §5-preferred shape in the abstract) and declined — zero occurrences in tree today, and it would touch FreeMonoid<PipelineStep> plus every Pipeline constructor for a currently-unexercised case (the purity trap, DESIGN.md §6). This is recorded as a typed note directly on the carrier: pipeline_steps_empty_arm_note beside Pipeline's declaration in src/v2/std/orchestration.dag, plus orch_if_empty_arm_note beside the fix in 05_emit_orchestration.dag — not just in this PR body, per DESIGN §6 "let the mark on the carrier be the authority."

Test plan

Four new discriminating RED witnesses in src/v2/workflow/orchestration_emit_test.dag, all asserting valid emitted bash (none asserting a refusal):

• orch_if_empty_then_no_else_noop_holds → "if [ -z \"$1\" ]; then :; fi"
• orch_if_empty_then_with_nonempty_else_noop_holds → "if [ -z \"$1\" ]; then :; else verdict=present; fi"
• orch_if_nonempty_then_with_empty_else_noop_holds → "if [ -z \"$1\" ]; then verdict=absent; else :; fi"
• orch_if_both_arms_empty_noop_holds → "if [ -z \"$1\" ]; then :; else :; fi"

Run via:

cargo run --release --bin claim_batch -- \
  --source-root src/v2 --source-root dag \
  --entry src/v2/workflow/orchestration_emit_test.dag \
  --function orch_if_empty_then_no_else_noop_holds \
  --function orch_if_empty_then_with_nonempty_else_noop_holds \
  --function orch_if_nonempty_then_with_empty_else_noop_holds \
  --function orch_if_both_arms_empty_noop_holds

All 4 PASS.

Byte-identity receipt (post-#7277 main)

Re-ran against main post-#7277-merge (this branch merged main via git merge, not rebase) to confirm the fix doesn't disturb any existing emitted bytes:

• All 5 pre-existing real If call-site witnesses (the 3 ci_floor_peak_emit.dag single-step sites covered by ci_floor_peak_emit_band_holds, and the 2 live_deploy/emit.dag no-else sites covered by live_deploy_emit_holds) — PASS, byte-identical.
• All 5 pre-existing If shape tests in orchestration_emit_test.dag (orch_if_single_step_no_else_byte_identical_holds, orch_if_single_step_with_else_byte_identical_holds, and the 3 multi-step block-form tests from Emit lane: block-bodied if_else row for multi-step then_/else_ bind #7277) — PASS, byte-identical.
• The 4 new empty-arm witnesses above — PASS.

Combined run (11 witnesses total), exit code 0, all PASS:

cargo run --release --bin claim_batch -- \
  --source-root src/v2 --source-root dag \
  --entry src/v2/workflow/orchestration_emit_test.dag \
  --function orch_if_empty_then_no_else_noop_holds \
  --function orch_if_empty_then_with_nonempty_else_noop_holds \
  --function orch_if_nonempty_then_with_empty_else_noop_holds \
  --function orch_if_both_arms_empty_noop_holds \
  --function orch_if_single_step_no_else_byte_identical_holds \
  --function orch_if_single_step_with_else_byte_identical_holds \
  --function orch_if_multi_step_then_no_else_block_form_holds \
  --function orch_if_multi_step_then_and_else_block_form_holds \
  --function orch_if_multi_step_else_only_block_form_holds \
  --entry src/v2/workflow/ci_floor_peak_emit_test.dag --function ci_floor_peak_emit_band_holds \
  --entry dag/test/claim/live_deploy/emit_test.dag --function live_deploy_emit_holds