Not really spotted by:	Coverity

The pkg-config file wasn't using the one defined by the configure
script and could lead to a mismatch if the package name changes.

I was wrong, or I lied. This is the final cleanup.

This time, once and for all.

Unlike HTTP/1 that relies on the underlying transport, we can time
streams out simply by not crediting them. This is our first usage of
the CANCEL error with the RST frame, even though this change is not
comprehensive since the window only applies to DATA frames.

In addition to the session send timeouts, we can now kill streams that
aren't making progress and free the resources they hold onto even when
the session is continuously active.

There is one caveat though, there is currently no timeout for trying
to write to the session, so unless a client maintains a stream forever
by not crediting it we won't catch it.

For that to happen the next step could be to change this pattern:

    H2_Send_Get(...);
    /* send something */
    H2_Send_Rel(...);

And instead check a return value:

    if (H2_Send_Get(...) == 0) {
        /* send something */
        H2_Send_Rel(...);
    } else {
        /* error handling */
    }

But unfortunately, not all uses of get/rel enclose an attempt to send
something.

Hopefully this has a newer version of brew that fixes the outdated error
we are currently experiencing.

on the workspace to check if any element of it is non-NULL.

Spotted by:	Steven

Spotted by:	fgs & ubsan

Instead of a timestamp per line, which makes diff(1) really hard
to use, emit millisecond timestamps on separate "dT" lines, whenever
time has changed since last timestamp.

so we apply it only where it makes most difference.

The hard work done by:	slink

Closes: 2792

That libtool bug on SunOS...

Spotted by VTEST.

This would trigger the tautological-compare warning on a recent-enough
Clang, and SunCC appears to complain as well.

Reported by @daghf.