Added support for propagating server timing metrics to client

[sigurdp]

Added two utility classes, AddProcessTimeToServerTimingMiddleware and PerfMetrics and that can be used to gather and propagate server timing metrics to the client.

Both these clases utilize the Server-Timing response header to communicate the chosen metrics.

The PerfMetrics class can also be used to gather metrics for logging in the backend logs.
Added example usage of PerfMetrics in backend/primary/routers/surface/router.py

[anders-kiaer]

Here we should at least already be sure int_duration_ms is of type int? 😄

[anders-kiaer]

I think __str__ would be more Pythonic:

Suggested change

	def get_as_string(self) -> str:
	return ", ".join([f"{key}={value}ms" for key, value in self._metrics_dict.items()])
	def __str__(self) -> str:
	return ", ".join([f"{key}={value}ms" for key, value in self._metrics_dict.items()])

Then it should be called automatically by Python when needing to convert the object instance to a string in various contexts. If we include ms-output as well in the generated string we can simplify lines like

    LOGGER.debug(f"Loaded property surface in: {perf_metrics.get_elapsed_ms()}ms ({perf_metrics.get_as_string()})")

which we probably will get many of since they are defined in each endpoint, to something simpler like

    LOGGER.debug(f"Loaded property surface in: {perf_metrics}")

which would also ensure we print out the performance metric instance in a consistent way across the app endpoints.

[sigurdp]

Went for a hybrid, to_string(self, include_total_elapsed: bool = True), that by default includes the total elapsed time and can be used directly in log statements.