bucky/model: WithFlashAttn(false) is ignored; flash attention always on #1041

Summary

bucky/model: WithFlashAttn(false) has no effect. Flash attention stays on, while the doc comment says it defaults to false and the startup log reports flash-attn=false.

Where

kronk v1.32.8 (also current main), sdk/bucky/model/model.go:59-67:

cp := whisper.ContextDefaultParams()
if cfg.UseGPU {
    cp.UseGPU = 1
} else {
    cp.UseGPU = 0
}
if cfg.FlashAttn {
    cp.FlashAttn = 1
}

• cp starts from whisper.cpp's defaults, and in whisper.cpp v1.9.5 those set flash_attn = true (src/whisper.cpp:3716, in whisper_context_default_params()).
• The FlashAttn branch only ever writes 1. Unlike UseGPU, it has no else branch, so false never reaches whisper.cpp.
• sdk/bucky/model/config.go:33-34: the Config.FlashAttn doc says "Defaults to false.", which is wrong in practice.
• model.go:101 logs "flash-attn", cfg.FlashAttn, which reports false while the kernel is active.

Repro

Load any whisper model with model.WithFlashAttn(false) (or with the option unset), using either WithUseGPU(true) or WithUseGPU(false). whisper.cpp logs:

whisper_init_with_params_no_state: flash attn = 1

Expected

WithFlashAttn(false) turns flash attention off, and the doc and log match what whisper.cpp actually uses.