--clock-mhz: report native-handler call counts at exit (fixes #109)

[sidick]

Native library handlers run in zero emulated cycles under --clock-mhz — documented, but previously invisible: a benchmark had no way to see how much of its work bypassed the cycle count. AmigaPorts/m68k-amigaos-gcc#89 hit this for real: GCC codegen patches A/B-benchmarked on a real A3000 vs volamos --clock-mhz 25 showed a memcpy benchmark improving 5.3% on hardware but 49.1% under volamos — the signature of copies vanishing into exec's native CopyMem.

• LibraryTable::dispatch counts every dispatched call per slot (one integer hashmap bump on the hot path; library/handler names are resolved only at report time) and accumulates the CopyMem/CopyMemQuick pair's D0 byte size, read before the handler runs so it's the caller's own value.
• Runtime::native_call_counts() exposes both, gated on a configured clock rate and scoped to the top-level run, exactly like emulated_cycles().
• The CLI prints a third report line under --clock-mhz (omitted when the guest made no library calls):

volamos: 183 native library calls ran in zero emulated cycles: exec.library/CopyMem x120, dos.library/Write x41, ...; CopyMem/CopyMemQuick moved 491520 bytes natively

Top 5 handlers named, the rest folded into "and N more"; the copy total is one combined clause so the pair's shared number isn't double-reported.

Non-goal (per the issue): billing approximate cycles for native handlers. This makes the unbilled work visible; a synthetic cost model is a separate discussion.

Tests: core coverage that dispatch counts CopyMem+CopyMemQuick and accumulates both calls' bytes, that the accessor is None without a clock rate; CLI formatting tests (top-N + tail fold, zero-copy clause omission, empty case). 1053 tests passing, clippy/fmt clean, strict docs build passes. Docs: CLI-Reference --clock-mhz section (example output + explanation + warning-bullet counterweight), --help text, Changelog 0.8.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GNNXMtYaZVHe8ihLL2GsrY