wallet: adopt chia_query::mempool_refusal and correct the is_definitive_rejection doc

[MichaelTaylor3d]

Task

Two coupled changes in dig-wallet, both landing on the next chia-query bump.

1. Adopt chia_query::mempool_refusal — do not keep a second copy

chia-query #50 moves the intrinsic-refusal classifier down into chia_query::mempool_refusal, byte-identical, including its exclusion rationale.

The list originates here: crates/dig-wallet/src/sage/chain.rs:241 BUNDLE_INTRINSIC_REFUSALS — 11 names, exact-match, ~60 lines of reasoning about what was deliberately excluded.

Once chia-query publishes it, re-export from there and delete the local definition. Two copies of this list is a rival implementation of a money decision: they would drift, and the drift decides whether a user's spend is reported refused or retried. The list moved down rather than being duplicated precisely so that cannot happen.

chia-query is currently pinned at 0.20.0 here while main is past 0.22.0, so this rides the version bump that closes that gap.

2. Correct the is_definitive_rejection doc

crates/dig-wallet/src/sage/rpc.rs:2326-2335 describes the old peer_then_coinset push routing. #50 changes that: coinset is never consulted on a refusal.

The new policy: an intrinsic refusal is final; anything else gets exactly one more peer; never more than two transmissions; the refuser is not ejected.

A doc asserting behaviour the code no longer has is a defect in its own right — it is what a future reader will build against.

Why the retry is safe, so the doc says the right thing

Two protocol facts, not optimism:

• an identical re-send shares the tx id, so it is not a double-spend;
• a peer answering NotAdmitted/Failed did not relay, so the second peer's verdict is clean.

The timeout-after-transmit race stays on the unchanged Err path.

A constraint that will bite if missed

TxStatus gets NO new field. dig-wallet builds it as a struct literal at chain.rs:876, so adding one breaks compilation here. The chia-query side was designed around that.

Blocked until

chia-query publishes the version carrying mempool_refusal (the #24/#27/#33/#34/#50/#56 batch). Release-first: the crate ships, then this adopts. Do not start before the version exists — the bump cannot resolve.

Also verify, do not assume

Whether anything else in dig-wallet reads the local list or depends on subscribe(capacity), which #34 removes in favour of source-scoped subscribe_frames(address, capacity). The measured future adopter of the fan-out is run_update_loop (sync.rs:1274), which follows exactly one session — consistent with the new shape, but confirm it rather than trusting this sentence.

Related

• push_tx treats one peer's NotAdmitted as final — decide whether a refusal deserves a second opinion chia-query#50 — the decision, in full
• peer: a subscription can be terminated by frames from a peer it is not following chia-query#34 — source-scoped subscriptions
• Parent: https://github.com/DIG-Network/dig_ecosystem/issues/2761

[MichaelTaylor3d]

Both halves of this ticket are being worked as part of the full chia-query 0.21.0 -> current adoption, tracked at #557 (parent epic https://github.com/DIG-Network/dig_ecosystem/issues/3194). The adoption PR will carry the Closes keyword for this issue.

Two measurements that update this ticket's own text, recorded here rather than left for the next reader to re-derive:

• The pin stated here is stale. This says "chia-query is currently pinned at 0.20.0 here while main is past 0.22.0". On origin/main (c0e3597) crates/dig-wallet/Cargo.toml:133 declares chia-query = "0.21.0", and current published is 0.24.0.
• The subscribe(capacity) question in "Also verify, do not assume" is answered: it is a no-op for this repo. dig-node contains zero references to FrameFanout, FrameSubscription or subscribe_frames. Its subscribe_store / unsubscribe_store are dig-node's own store subscriptions, unrelated to the frame fan-out. run_update_loop (sync.rs:1274) is a future adopter, not a current caller, so the 0.24.0 removal breaks nothing here.
• TxStatus still gets no new field. Verified against chia-query v0.24.0, so the struct literal at chain.rs:876 is safe.

Sequencing note: the pin will not be 0.24.0. That version carries DIG-Network/chia-query#59 (HIGH), a drive-loop wedge that freezes the coin-state cache with no rearm signal — precisely the failure a wallet must not adopt. The code work proceeds now; the version is taken after #59 publishes.