Metrics are not grouped by type

[marksilcox]

Describe the bug
When calling the metrics endpoint I would expect them to be grouped by type. However, some metrics are grouped under the incorrect type header

To Reproduce
Steps to reproduce the behavior:

1. Run Pulsar v2.6.0
2. Fire some messages to topics
3. Access metrics endpoint
4. See error

Expected behavior
All metrics are grouped appropriately.

Screenshots

Desktop (please complete the following information):

• OS: Docker on MacOS, Amazon Linux on AWS

[codelipenghui]

@marksilcox The broker just expose the Prometheus formatted metrics, if you want to do some aggregate calculations, it's better to achieve at the Prometheus.

[marksilcox]

@codelipenghui I am not wanting to aggregate. As you can see from the screenshot, some metrics are not being grouped under their # TYPE header. See # TYPE pulsar_msg_backlog gauge header, under that I can see 9 other metrics not related to that header. This happens in a few other places and metrics are scattered across the output.

Prometheus docs states "All lines for a given metric must be provided as one single group" - https://github.com/prometheus/docs/blob/master/content/docs/instrumenting/exposition_formats.md#grouping-and-sorting