v0.14.1

Fixed

• Corrected the benchmarking caveat published in 0.14.0. It attributed the earlier bad measurement to an on-access virus scanner rescanning build outputs on every exec, citing a 5.0x penalty (306 ms → 61 ms). That does not reproduce — the same interleaved comparison on a settled tree measures 1.03x. The original was taken while a concurrent session was rebuilding oam, so every exec hit different bytes. The guidance survives with the right reason: measure an installed oam, never one out of target/, because a build directory's contents change underneath a running benchmark. The performance figures in 0.14.0 were re-verified against an installed binary and are unchanged — node 167 ms vs oam 112 ms, 0.67x.