refactor!: pass Target to BnbMetric methods instead of storing it

[evanlinjin]

Motivation

BnbMetric implementations currently each store their own Target (LowestFee { target, .. }, Changeless { target, inner }). That has two problems:

1. Changeless<M> duplicates the target and must keep its copy in sync with the inner metric's — a silent-mismatch footgun (both fields are public and set independently).
2. It's inconsistent with the rest of the library, where target is always a parameter: cs.excess(target, drain), cs.is_target_met(target), cs.drain(target, policy). The metric structs are the only place target is baked into state.

What this does

Make target a parameter of the metric methods (and the BnB entry points):

pub trait BnbMetric {
    fn score(&mut self, cs: &CoinSelector<'_>, target: Target) -> Option<Ordf32>;
    fn bound(&mut self, cs: &CoinSelector<'_>, target: Target) -> Option<Ordf32>;
    fn drain(&mut self, cs: &CoinSelector<'_>, target: Target) -> Drain;
    fn requires_ordering_by_descending_value_pwu(&self) -> bool { false }
}

// cs.run_bnb(target, metric, max_rounds)
// cs.bnb_solutions(target, metric)

Consequently:

• LowestFee → { long_term_feerate, dust_relay_feerate, drain_weights } (no target).
• Changeless<M> → Changeless<M>(pub M) — with target gone it collapses to a single field, so it becomes a plain newtype: Changeless(LowestFee { .. }). The duplicate target is gone by construction, so the "must match inner's target" invariant is deleted rather than documented.
• BnbIter holds the single target and threads it to the metric.

Conceptually this reflects the right split: target is the problem/constraints; the metric is the objective. The feerates/drain-weights stay in the metric (objective config); target — universal and problem-level — becomes a shared input, so a mismatch across composed metrics is unrepresentable.

Argument order is cs, then target (then the rest), keeping the same cs-before-target relative order the library already uses (where cs is usually the receiver).

Notes

• The target field and its plumbing simply disappear, so the net diff is small despite the wide surface.
• Breaking change on the unreleased 0.5.0 API surface (post-feat!: BnB metrics decide the change output themselves #49).
• cargo fmt / clippy / doc / full test suite + doctests / no_std build are all green.

🤖 Generated with Claude Code

[evanlinjin]

ACK 304ac2d