Make mining benchmarks measure full nonce ranges

[n13]

Overview

Mining throughput benchmarks could include context-construction overhead or stop early when a low-difficulty target produced a solution. This makes measured hash rates less representative of steady-state mining.

What changed

• Use U512::MAX difficulty (target 1) for CLI, Criterion, and WGPU hashrate throughput measurements.
• Build benchmark job contexts outside timed Criterion loops.
• Make inclusive nonce ranges contain exactly their labeled count.
• Remove the deliberately early-exit solution-finding case from the throughput suite.
• Print the effective benchmark difficulty in CLI output.

Validation

• cargo fmt --all -- --check
• cargo clippy -p miner-cli -p engine-cpu -p engine-gpu --all-targets --locked -- -D warnings
• cargo check -p engine-cpu --bench cpu_engine_bench --locked
• cargo check -p engine-gpu --bench gpu_engine_bench --example hashrate --locked
• cargo test -p miner-cli --locked
• cargo run -p miner-cli --locked -- benchmark --cpu-workers 1 --gpu-devices 0 --duration 1

Risks and mitigations

The Criterion suite no longer measures solution-discovery latency. That case mixed variable early exits with throughput measurements; functional solution discovery remains covered by engine tests.

Follow-ups

None.

[n13]

Don't need this!