doc: fix multiple Doxygen and breathe problems

[gmarull]

Introduction

This PR is an attempt to fix multiple documentation problems we have when using Doxygen + Breathe.

Summary of the changes

• Disabled INLINE_SIMPLE_STRUCTS Doxygen option. Some struct pages were missing in the Doxygen output (e.g. check https://docs.zephyrproject.org/apidoc/latest/group__bt__gatt__client.html)
• Removed a duplicate definition in ieee802154 headers. While valid C code, as of today Sphinx C domain/breathe do not support duplicate definitions. In this case I think the duplicate was likely unintentional.
• Doxygen (and subsequently Sphinx/Breathe) got confused by the ATOMIC_DEFINE inside a struct (see https://docs.zephyrproject.org/apidoc/latest/structbt__gatt__subscribe__params.html). Refactored this part.
• Introduced a new Sphinx extensions to handle warnings natively, so without the need of post-processing the output.

Remaining issues

The remaining issues should be fixed by sphinx-doc/sphinx#8313. TLDR; as of today you can't have something like:

 /* demo considered a duplicate because the name is used by both struct and function */
struct demo {
...
};

void demo(...);

/* id considered duplicate */
union {
	uint16_t id;
	struct {
		uint16_t id;
	} vnd;
};

After Sphinx fixes the problem above we should be able to (hopefully) get rid of the current "known issues".

zephyr.warnings_filter extension

Instead of post-processing Sphinx output, I have created a small extension that natively filters out Sphinx warnings. This Sphinx plugin can be used to filter out warnings that are known to be false positives. The warnings are filtered out based on a set of regular expressions given via a configuration file, the same principle as before.

At the same time I have simplified current regular expressions and moved them to doc/known-warnings.txt with the idea to make docs more self-contained.

[jukkar]

LGTM

[carlescufi]

API meeting 2021.04.06

• Remove the commit net: net_ip: hide some external and forward declarations from Doxygen. Instead of ifdefing out the forward declarations, the warnings will be suppressed by the new plugin
• Otherwise agreed to the changes

[marc-hb]

While testing #33448 I noticed that stderr output from Sphinx (and more commonly) from the very many subprocesses it forks) escaped the defunct filter-known-issues.py and tended to get drowned in the rest of the output and missed. @gmarull I'm curious whether what is the current status of stderr after this PR, unchanged? Just wondering whether you're aware of that since you're on top of everything documentation-related now.

[gmarull]

While testing #33448 I noticed that stderr output from Sphinx (and more commonly) from the very many subprocesses it forks) escaped the defunct filter-known-issues.py and tended to get drowned in the rest of the output and missed. @gmarull I'm curious whether what is the current status of stderr after this PR, unchanged? Just wondering whether you're aware of that since you're on top of everything documentation-related now.

This PR introduced a way of filtering warnings that does not deal anymore with the output of Sphinx. What it does is to directly intercept Sphinx warning logs and filters them out (or converts them to INFO in non silent mode). So the stderr will be what Sphinx originally outputs there.