Fix test failures on newer kernels

[tklauser]

Commit 1:

Enable extended error message reporting in tests

This causes the NETLINK_EXT_ACK socket option to be set and extended
error messages to be reported in errors. This should help debug test
failures.

Also use errors.Is to check for errors because using extended error
reporting, the errors no longer match exactly.

Commit 2:

nl: avoid trailing NULL byte in error messages

Use unix.ByteSliceToString to convert the NULL-terminated
NLMSGERR_ATTR_MSG error message.

Commit 3:

Use valid Tos value in TestRouteFilterAllTables and TestRouteExtraFields

TestRouteFilterAllTables and TestRouteExtraFields started failing a
while ago after GitHub actions images bumped the kernel version from
5.15.x to 6.2.x [1].

This is because newer kernels containing commit [2] started rejecting
Tos values with the ECN bits set to non-zero, this leads to the
following error being reported when adding routes:

    invalid argument: Invalid dsfield (tos): ECN bits must be 0

Fix this by using a valid Tos value in TestRouteFilterAllTables and
TestRouteExtraFields.

[1] https://github.com/vishvananda/netlink/pull/773#issuecomment-1754436653
[2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f55fbb6afb8d

Commit 4:

Use inbound policy in TestXfrmPolicyWithOptional

Since kernel commit
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=3d776e31c841b
optional tunnel/BEET mode templates in outbound policies are rejected.

Use an inbound policy instead to fix the test failure on newer kernels.

[tklauser]

@vishvananda @aboch PTAL. This should fix the CI failures currently seen on the main branch after GitHub actions runner images updated kernel versions, also see #773 (comment)

[aboch]

Thanks @tklauser for your fixes!

Usually we want a commit per PR, but in this case the commits are clearly fixing/improving independent things. Since all changes pertain to UT fix, I will not ask for four independent PRs.

LGTM

[tklauser]

Usually we want a commit per PR, but in this case the commits are clearly fixing/improving independent things. Since all changes pertain to UT fix, I will not ask for four independent PRs.

Thanks, I wasn't aware of that rule. I'll keep it in mind for future changes.