fix(bench): measure native-binary size from the published npm artifact

[BryanFRD]

The "Native binary" size on the performance page came from the runner's local cargo build --release binary, while the npm size and all competitors come from a real npm install of the published package — so they weren't comparable (and the local binary measured larger than the npm install that contains it).

Measure the native binary from the published binary inside the npm package (@ferrflow/<platform>/bin/ferrflow), installed the same way as the npm footprint. Now both sizes describe the same published release, the binary is a subset of node_modules (so binary ≤ npm), and the column reflects what users actually run. Dead get_binary_size removed.

Validated on the next benchmark run (graceful N/A if the platform package is absent). bash -n clean; the bats suites test ferrflow_binary_size_mb as an input field and are unaffected.

Closes #149