fix(scoring): reject sub-dust-only rates in is_executable_rate

[anderdc]

Problem

is_executable_rate is the crown-eligibility gate (scoring.py executable_check). It only checks that some positive integer source ≥ 1 maps a TAO leg into [min_swap, max_swap]. At the live SN7 bounds (0.1–0.5 TAO) that lets a rate whose only in-bounds source is a single satoshi through as "executable":

• btc→tao at 5e7 TAO/BTC → 1 sat → 0.5 TAO
• tao→btc at 2e-8 TAO/BTC → the inverse of the same boundary (swap 1670)

A 1-sat BTC output is below Bitcoin's dust limit (P2WPKH dust = 294 sat, bitcoin/policy.cpp), so it can never be broadcast. The rate is therefore unfundable, yet today it:

1. Earns crown — these boundary rates win the rate sort and collect reward despite being un-routable.
2. Powers a grief — a reservation taken against one forces an unfulfillable swap that times out and slashes the miner (swap 1670), and on btc→tao the timeout pays the reserver the full TAO leg for a sub-dust BTC input.

Fix

Floor the routable-source search at a per-chain min_onchain_amount (the smallest amount that can actually exist/move on-chain):

• BTC: 1000 sat — verified transferable: IsDust uses strict < and 546 is the P2PKH dust line, so 1000 clears every standard output type with margin against higher dustrelayfee and wallet quirks. Also tightens the executable-rate ceiling to TAO_max/(10·1000) = 50,000 TAO/BTC (~171× market).
• TAO: 500 rao — the existential deposit. Inert today (_has_integer_routable_source is only ever called with the non-TAO side), so TAO behavior is unchanged; defined for future chains.
• default: 1 — no floor; any future chain just declares its own.

One line in _has_integer_routable_source: max(1, …) → max(src.min_onchain_amount, …). The existing inverse-symmetry path means the same change rejects both crown-squat directions.

Effect (and honest scope)

• Sub-dust-only rates now fail executable_check → dropped from crown qualification, and the surviving rate band becomes fundable, so an honest arbitrageur can finally reserve-and-punish an absurd rate (reserving makes the miner busy, which ends its crown, and forces deliver-or-slash). This is the breadth/depth doc's "allow extreme rates only if they can be backed up against a real reserver" — min_sat is the prerequisite that makes "backed up" physically possible.
• Does not by itself stop crown-squatting: the executable ceiling moves to ~50,000 TAO/BTC, so a squatter can re-post just under it (now fundable, hence arbitrage-exposed). Eliminating the squat is the breadth/depth redesign (rank spillage + EMA quality bonus + a σ-gate on the posted rate), tracked separately.
• Does not gate the reservation path (axon_handlers still only checks slippage). Gating reservations by is_executable_rate is a planned follow-up that closes the dust self-park at the source.

Deferred (consciously)

• max_sat / dest-dust check — the sub-dust destination edge on absurd high tao→btc rates (a crown-loser, never triggers at market rates). Hygiene.
• is_executable_rate at the reservation path.
• Breadth/depth scoring + reservation bid window — the real crown-squat / arbitrage-swoop fixes.

Tests

• Rewrote the two tests that asserted the old "1-sat boundary is executable" behavior to pin the dust-floor boundary (sub-dust rejected, dust-clearing boundary executable), in both directions; they derive the floor from min_onchain_amount so they track the constant.
• Full suite: 617 passed.