Optimize performance by removing external processing when making calls to sdk

[plengauer]

external processes are used because handles are managed by the SDK. a pipe needs creating, before that a random file name, then it needs to removed afterwards. echo and cat are builtins. if the sdk manages the handles, almost never information needs to be communicated back, except for traceparents.

the handle must be unique in the sdk, but with subshells spawning different processes but reusing sdks, that may be an issue. to avoid that, we could use the REAL pid. $$ is not the real pid, but always the pid of the toplevel shell. we could use tricks like read -r /proc/self/stat to really find our own process id in non bash shells where $BASHPID is not an option

[plengauer]

this seems to be much harder because of implicit subshells. because even on usage level, it will look like "$(otel_span_start ...)" and it will always be in a subshell.