Clean remaining prose-comment rows: bilateral seed deletes + typed Disposition conversions

[gunbai-bot]

Third pass on the comment-smuggled-in-strings habit (operator: "could you clean the rest?"). #6424 deleted the dead rows, #6428 converted the grep-witnessed ones; this PR takes the classes previously excluded.

What this does

• 25 seed-realized prose rows deleted bilaterally — the .dag data row AND its generated stage0 pub fn (every one had zero callers on both sides). src/v1/{04_env,04_infer,04_resolve,05_emit_rust,dag_collect,dag_collect_support}.dag, dag/std/{algebra,computation,measure,realization_schedule}.dag, and their std_*/v1_compiler_* seed realizations.
• 6 rows converted to typed Disposition = Scaffold: ci_witness_optin_inversion (the CI opt-in inversion dissolve-mark), export_signature_facts_host_scaffold_dissolution_trigger (+ firewall-test grader rewritten to a typed match), interface_summary_v0_dissolution_trigger, host_kernel_relu_mul_add_op_codes_scaffold_note, and the hand_lens_host_bridge scaffold-index pair — including typing the watchdog's dissolution_trigger field String → Disposition.
• 1 duplicate deleted: tailscale_acl_phase2_design_status (a typed live-write disposition row already existed; the witness's prose greps dropped, typed check kept).
• 5 dead notes deleted in the frontier-resolvable src/v2/test/manual files.

Known-blocked residue (1 row)

rust_literal_emit_string_refusal_note stays prose: importing std.disposition into its closure trips the variant-name collision wall — Terminal is declared by both GrammarExpr and Disposition — the exact cross-tree duplicate-name hazard the forensics_6280 report documented. Unblocks when that §3 name fork is resolved.

Also still standing, per earlier rulings: 3 notes inside frontier-unresolvable forensics/ownership fixtures (pending operator ruling on deleting those fixture files wholesale), the ci.yml-emitted doc notes (comments in a generated artifact are a legitimate medium), and genuine prose-shaped data (prompts, templates, expected outputs).

Verification

• Every diff .dag file entry-probes clean (src/v1 files excluded — outside the frontier roots, proven by Sweep dead prose data-String rows (comments smuggled in strings): 215 rows across ~130 files #6424's green run touching src/v1/dag_collect.dag).
• Watchdog, tailscale phase-2, and interface-summary firewall witnesses green by execution.
• Seed: cargo check -p v1-compiler green locally. Correction to the commit message: it claims nextest ran locally — it did not (nextest isn't installed in this container); the full test suite gate is the PR's rust_tests job.

[gunbai-bot]

Local seed-suite receipt (completing the verification note in the PR body): cargo test -p v1-compiler-tests → 554 passed, 2 failed, 54 ignored. Both failures are local environment preconditions, not regressions from the bilateral seed deletions: gunbhub_serve_program_runs_green panics at its precondition release gunbc binary required and claim_batch_hermetic_flag_uses_mock_response_by_execution at claim_batch binary missing — CI's rust_tests job builds both binaries before running, so the authoritative gate is this PR's rust_tests check.

— sent from smart-boar-596