feat(wallet)!: move the reservation codes off -32044, which dig-node already owns

[MichaelTaylor3d]

Outcome

-32044 meant two opposite things across the repo boundary. The contract moves; dig-node does not.

code	symbol	disposition
-32044	WALLET_NODE_SPEND_DISABLED	terminal — registered here with dig-node's semantics
-32046	WALLET_COINS_RESERVED	transient wait — was -32044
-32047	WALLET_RESERVATIONS_UNAVAILABLE	unknown — was -32045

This reallocation is deliberately NOT additive, and should not read as a routine bump: the
numeric wire codes of two codes published in 0.20.0 change. That is the right trade because 0.20.0
is ~40 minutes old with zero implementors (grepped: neither symbol nor either numeric appears in
dig-node or dig-app, both of which pin 0.17/0.19), while dig-node's -32044 is live on every
default install and its refusal is a custody decision — the node relays bundles others signed,
but spending its own coins is default-OFF, because a caller could otherwise sign through the node
and hand the bundle straight back. Leaving -32044 ambiguous is worse than renumbering two codes
nothing implements.

-32045/-32046/-32047 are all unused in dig-node (checked against every -320xx literal in
dig-node/crates/); -32045 is vacated only so the pair stays adjacent as one coherent block.

WALLET_NODE_SPEND_DISABLED's wording is copied from dig-node SPEC.md:3027 so the two catalogues
agree verbatim on the remedy.

The root cause, and the durable half

dig-node minted a code inside the shared -3204x range that was never declared in the shared
catalogue, so nothing could see the clash at allocation time.

• every_catalogued_code_is_numerically_unique — pairwise across the whole catalogue, plus a
  from_code round-trip. The pre-existing per-variant assertions only compared ONE code against the
  rest, so a collision between two codes neither of which was under test was invisible to them.
• a_terminal_custody_refusal_and_a_transient_wait_never_share_a_code — pins each number
  together with its disposition, so a future re-collapse is visible. Asserting only that the
  numbers differ would pass for a catalogue that had quietly given both variants the same meaning.
• The range-ownership rule is now normative in SPEC.md §5 and the error.rs module docs: the
  -3204x band is owned by this document, and a node or client MUST NOT mint into it privately.
• SPEC.md also no longer claims "all four wallet codes mean the answer is UNKNOWN" — false since
  0.20.0 and doubly so now. It states per-code dispositions and requires branching on the symbol,
  not the band.

Guard proof (a uniqueness test that passes against the colliding version proves nothing). With
WalletRateLimited moved to -32044, the named test fails with the exact clash:

thread 'kats::every_catalogued_code_is_numerically_unique' panicked at src/kats.rs:1965:
assertion `left != right` failed: WALLET_RATE_LIMITED and WALLET_NODE_SPEND_DISABLED both claim -32044
test result: FAILED. 0 passed; 1 failed; 151 filtered out

Reverted; tree restored to the committed state before finalizing.

Expressibility, and what dig-node owes

Only half the guard is expressible here. This crate cannot see dig-node's sources, so uniqueness
within the catalogue is all a test here can assert; the cross-repo half is the range-ownership
rule, which only dig-node can honour. dig-node owes the reciprocal check — assert every
-3204x it mints is declared in this crate's ControlErrorCode::ALL, which it can express because
it already depends on this crate. Related finding (comment, not a ticket): dig-node also mints
-32033 CONTROL_INGRESS_LIMITED and -32060 PEER_PING_REFUSED, plus -32001..-32017 and
-32050..-32052, none declared here. Those are outside -3204x and none currently clash, so they
are not fixed in this PR.

Blast radius checked

Grep-based (gitnexus fallback, per §2.0 bound 2 — a ~10-minute per-worktree analyze is not
justifiable on a lane two others are blocked on, and the radius here is fully enumerable). Targets:
ControlErrorCode::{code,name,origin,description,ALL} and the two moved variants. Every reference
in-crate: error.rs, kats.rs, and doc-links in params.rs/results.rs/traits.rs/method.rs
(symbolic only, no numerics). Out of crate: zero — no consumer references either symbol or
either numeric.

Verification

• cargo test --all-features — 152 lib passed, 9 doc passed, 0 failed (baseline 150 + 9; +2 are
  the new guards). Parsed from the test result: lines, one cargo process at a time.
• cargo clippy --all-features --all-targets -- -D warnings — clean.
• §2.4b is a no-op, confirmed not assumed: Cargo.toml declares no dig-* and no chia-*
  dependency (only serde/serde_json/semver/async-trait/futures); there is no
  package.json, so no @dignetwork/* deps either.

Version

0.20.0 → 0.21.0, titled feat(wallet)!:. Note this crate's cliff.toml does not render
breaking as breaking (#26), so the ! is for git log and this body, not the changelog.

Progress: https://github.com/DIG-Network/dig_ecosystem/issues/3127