fix(cli): show typed graph receipt recovery fields (FUZZ-039 r2)

[randlee]

Fixes comp-t-7.14.1 (sanity finding on FUZZ-039, comp-t-7.14): human-mode graph refusals print the typed error's structured details (affected id, read cause) and recovery guidance. Adds an additive execute_bead_request_with_diagnostics callback API in sc-composer-beads (existing entrypoints delegate with a no-op; receipt JSON, signatures and the Python binding unchanged). Tests run against a fake bd so they always execute.

Rebased 5541415 -> ad1c983 onto #717. Gates at ad1c983: clippy -D warnings clean, fmt clean, cargo test --workspace 1066 passed / 0 failed / 0 ignored.

🤖 Generated with Claude Code

[randlee]

QA fix verification: comp-t-7.14.1-qa (D1-F1 / FUZZ-039) at ad1c983

Verdict: PASS. 1 carried finding verified fixed.

Finding	Disposition
D1-F1	fixed

The fuzz coordinator, locked to D1-F1, verified at the pinned commit through the real CLI against a fake-bd stub:

• print_human_receipt (via run_bead collecting diagnostics only when not --json) now prints the real error message plus details: (id and cause) and recovery: lines. Exercised: BEADS_GRAPH_PARENT_NOT_FOUND, BEADS_GRAPH_RELATION_INVALID, BEADS_GRAPH_ID_INVALID, BEADS_GRAPH_CONFLICT, BEADS_GRAPH_READ_FAILED. Read only, not run: SCOPE_MISMATCH, FORMULA_UNSUPPORTED, unknown relation endpoint, EDGE_MISSING (same generic diagnostics path).
• The printed text matches the real error.rs message ("graph parent nosuch does not exist; ..."); the old test wording is gone.
• Two regression tests (fuzz_039_human_preview_attach_prints_parent_refusal_reason, fuzz_039_human_receipts_preserve_conflict_id_and_read_cause) are un-ignored, assert message, details and recovery, pass (fuzz_039: 2 passed) and both fail on the base origin/sprint/t-7-phase-fuzz.
• The --json branch is unchanged in code; the manual text matches the printed form. Human output of success receipts and the attach auth refusal is byte-identical to baseline.
• Local fmt, clippy -D warnings and tests are clean at this head.

Notes, not findings of this round:

• Control and bidi characters in ids reach the terminal raw in the human message line (ESC in the message line, U+202E in message and details); this is FUZZ-040 territory (round 2 poured).
• GRAPH_READ_FAILED prints the cause twice (raw stderr excerpt plus message). Cosmetic.
• The same commit shell-quotes the bd dep add recovery commands for EDGE_MISSING (a human-output change; the manual says only that they are printed). The unit test for it was not run by the verifier.

Disclosures: --json outputs for the other cases are not byte-identical to baseline, but the differences come from other commits in this stack (temp input path, --json -- argv ordering, elapsed_ms, the bad-id code now BEADS_GRAPH_ID_INVALID); the FUZZ-039 commit was not isolated, so that conclusion rests on reading the diff. The carried finding was filed by dev-sanity, not a QA reviewer, so the coordinator was hand-dispatched (fix-round-scope accepted sc-sanity-llm as owner, but it is not a QA reviewer). CI not checked (rate-limit rule).

[randlee]

QA fix verification: comp-t-7.14-qa (FUZZ-039) at ad1c983

Verdict: PASS. 1 carried finding verified fixed.

Finding	Disposition
fuzz-039	fixed

This task checks the same PR and the same pinned commit (ad1c983) as comp-t-7.14.1-qa, which I closed PASS minutes earlier after a full verification by the fuzz coordinator. I did not dispatch a second verifier; the evidence from that run covers both deliverables of comp-t-7.14:

• Fix: in human mode print_human_receipt now prints the real error message plus details: (id and cause) and recovery: for the graph refusal codes (parent not found, relation invalid, id invalid, conflict, read failed exercised through the real CLI against a fake-bd stub; scope mismatch, unsupported formula, unknown endpoint and edge-missing read only). --json code path unchanged; manual text matches.
• Regression tests: fuzz_039_human_preview_attach_prints_parent_refusal_reason and fuzz_039_human_receipts_preserve_conflict_id_and_read_cause are un-ignored, assert message, details and recovery, pass, and both fail on the base origin/sprint/t-7-phase-fuzz. The test from comp-t-7.14's own commit (t-7 FUZZ-039: regression test for human-mode graph refusal details #714) was the one the dev-sanity checker had faulted for a wrong expected message; the message now comes from the real error.
• Local fmt, clippy -D warnings and tests clean at this head.

Same notes and disclosures as the 14.1 report on this PR (raw control/bidi characters in human text belong to FUZZ-040 round 2; --json byte-identity not isolated to the FUZZ-039 commit; fake bd; CI not checked per the rate-limit rule).