Disposition carrier slice-2: ratchet disposition_redundancy lens over the post-wall data:String marker fleet — migrate each region (bytes_seam, anthropic closure rows, budget-tree leaves, #5605 consumed_input_closure rows, unit_must_run_staged_note) from prose data:String to typed Disposition fields

[briansrls]

Disposition carrier slice-2: migrates the post-wall data:String marker fleet to typed Disposition fields and wires all regions into the disposition_redundancy lens test.

What changed

Five regions migrated from `data:String` prose to `data:Disposition` (Terminal or Scaffold):

Region 1 — bytes_seam (`dsl/std/bytes.dag`, `dsl/std/encoding.dag`)

• `bytes_seam_host_realization_marker: Disposition = Scaffold{bind: std.bytes.builtin_function_registry}`
• `utf8_decode_bytes_host_realization_marker: Disposition = Scaffold{bind: std.bytes.builtin_function_registry}`
• `base64_group_fill_carrier_debt_marker: Disposition = Terminal{debt paid: closed-sum enums already declared}`

Region 2 — anthropic closure rows (`dsl/extdeps/llm/anthropic.dag`)

• 4 `List` rows → `List`: two residual/coverage-gap lists, one serde receipt (all Terminal), one file-image beta list

Region 3 — budget-tree leaves (`dsl/gunbc/ci_floor_measurement.dag`)

• 3 new `data:Disposition` rows added (these had dissolution-trigger comments deleted in Parser-wall: remove DAG comment trivia rules (fail-closed by construction) #5579 and were never re-added as String rows):
  • `gunbc_ci_floor_measured_peak_disposition` → Scaffold binding `gunbc_ci_job_peaks` (EXISTS, fires correctly — live redundancy signal)
  • `gunbc_ci_host_fixed_overhead_disposition` → Scaffold binding planned `gunbc_ci_per_service_overhead` (silent)
  • `gunbc_ci_concurrent_runs_conservative_high_disposition` → Scaffold binding `runner_deployment_plan` (EXISTS in gunbc.ci_runner_placement, fires correctly)

Region 4 & 5 — #5605 consumed_input_closure + unit_must_run (`dsl/tools/rust_stage0_gates.dag`)

• 6 `data:String` rows → `data:Disposition`: 3 Terminal (permanent §5 invariants) + 3 Scaffold (planned successors that don't yet exist → lens stays silent until they land)

Test wiring (`src/v2/lens/disposition_redundancy_test.dag`, `src/v1/tests/src/disposition_redundancy_lens_test.rs`)

• Each region adds: nonempty probe (red-on-revert) + green-against-nonmatching + discriminating fire probe
• Rust harness extended with by-execution assertions for all new regions

Why

Closes out the post-#5579 data:String backlog: `//` comments were banned as a parse error in #5579, so prose dissolution triggers became untyped String rows. Slice-2 promotes them to `Disposition`, making the Terminal/Scaffold distinction structural and lens-checkable.

Faithfulness note (for review)

`structural_coverage_gap_anthropic_messages_200_residual` item 5 (the AnthropicBetaContentBlock Scaffold) collapses 9 specific pending variant names from the original prose:
`variant_pending: web_fetch_tool_result | code_execution_tool_result | bash_code_execution_tool_result | text_editor_code_execution_tool_result | tool_search_tool_result | advisor_tool_result | mcp_tool_use | mcp_tool_result | container_upload`

The collapsed Scaffold binds to `AnthropicBetaContentBlock` (doesn't exist yet). The dropped enumeration is recoverable from the Anthropic spec (case a from the parent coordination message) — these are precisely the block type names from the Anthropic beta API, no internal provenance. When `AnthropicBetaContentBlock` is declared, the variant modeling belongs in its own modeling TODO, not here.

Bind target corrections vs previous commits: items 2 and 3 in `structural_coverage_gap_anthropic_messages_200_residual` were originally bound to `AnthropicToolDefinition` and `AnthropicServerToolName` (both EXISTING types — lens correctly fired). Fixed to:

• Item 2: `MessagesToolUseBlock.NamedField{caller_payload}` — dissolves when `caller: Json?` is typed
• Item 3: `AnthropicServerToolUseSurface` — dissolves when a coproduct keyed by `AnthropicServerToolName` is declared

Ordering note

This PR lands BEFORE slice-3 (#5632, cool-carp). Slice-3 does a tree-wide `DeclLocator → std.decl_ref.DeclarationRef` sweep that covers the new bind targets added here. Do NOT switch to `DeclarationRef` in this PR — the type doesn't exist yet; cool-carp rebases over this merged PR and sweeps.

Test plan

```
ctrl-build -- cargo test -p v1-compiler-tests disposition_redundancy --lib
```

Result: `test disposition_redundancy_lens_test::disposition_redundancy_lens_discriminates_by_execution ... ok` (1 passed)

No pre-existing test failures introduced; `rust_stage0_gates_witness_test.dag` only imports the functions (unchanged), not the converted data rows.

[gunbai-bot]

All 6 CI failures are pre-existing on main, not introduced by this PR.

Root cause: PR #5647 ("Converge construction_justification onto std.disposition", merged to main) added import std.disposition { SingleAuthority } to multiple src/v2/lens/*.dag files (affected_set.dag, application.dag, affected_set_examples.dag, common/construction_justification.dag, and others). The witness_option_bridge_test and several lens tests use a source-root that only includes src/v2, not dsl, so they cannot resolve std.disposition (which lives at dsl/std/disposition.dag).

Verification: All 6 failing tests reproduce identically on the current main HEAD (no diff from this branch):

• witness_option_bridge_test::mark_excluded_no_longer_pattern_match_fails → same unresolved import: module 'std.disposition' not found error
• fact_cardinality_lens_test, extdeps_shape_transport_policy_lens_test, languages_consumer_census_lens_test → same failure
• pipeline::github_token_* and pipeline::github_create_review_* → same failure

This PR adds only dsl/std/disposition.dag (the carrier), dsl/std/bytes.dag + dsl/std/encoding.dag + dsl/gunbc/ci_floor_measurement.dag + dsl/tools/rust_stage0_gates.dag typed rows, and src/v2/lens/disposition_redundancy*.dag/rs test coverage. None of those touch the source-root configuration for witness_option_bridge_test or the affected lens tests. The pre-existing break is owned by the #5647 lane.

— sent from calm-otter-582

[gunbai-bot]

The conflict markers finding was valid at the prior head (the 1ef1f6669 merge had committed unresolved markers). Fixed in fc3eee910:

• Reset dsl/gunbc/ci_floor_measurement.dag to origin/main (drops the stale DashboardSessionReservation rows superseded by Unified conserving fleet-resource authority: CI runners + dashboard sessions derived together, fail-closed #5650)
• Re-applied only the genuine slice-2 additions: 1 import line + 3 Disposition data rows

Verifications on the current head:

• (a) git grep -nE '^(<<<<<<<|=======|>>>>>>>)' dsl/gunbc/ci_floor_measurement.dag → ZERO
• (b) gunbc compile --source-root dsl --source-root src/v2 → 518 files, 0 diagnostics
• (c) git diff origin/main...HEAD --stat → 6 disposition-only files

CI is now queued on fc3eee910 (run 28056820415).

— sent from calm-otter-582