chore(call): suppress mpg123 buffer underflow warnings in npm run call

The `speaker` npm package delegates to mpg123/coreaudio on macOS, which
writes `Didn't have any audio data in callback (buffer underflow)` lines
directly to stderr from native C code whenever the playback queue
briefly empties between WebSocket audio chunks. These are normal
operational noise (not errors) but can dominate the CLI output 3-15
times per call.

Wrap the `call` script with `bash -c` + a stderr process substitution
filter that drops any line containing "buffer underflow". The wrapper
uses `exec` so signals (Ctrl+C) pass through unchanged to tsx. bash is
invoked explicitly so the filter works regardless of npm's default
script-shell (sh/dash on some Linux distros doesn't support process
substitution).

Only affects the `npm run call` CLI — no runtime behavior change for the
voice agent or any other command. Every downstream customer clone of
this template benefits automatically.

Author: dhruva-vapi

package.json

@@ -10,7 +10,7 @@
     "apply": "tsx src/apply-cmd.ts",
     "push": "tsx src/push-cmd.ts",
     "pull": "tsx src/pull-cmd.ts",
-    "call": "tsx src/call-cmd.ts",
+    "call": "bash -c 'exec tsx src/call-cmd.ts \"$@\" 2> >(grep --line-buffered -v \"buffer underflow\" >&2)' --",
     "cleanup": "tsx src/cleanup-cmd.ts",
     "eval": "tsx src/eval.ts",
     "build": "tsc --noEmit",