This is to avoid a confusion between cancelling a fetch from a client
transaction and abandoning a fetch from a backend transaction.

Spotted by: @dridi

Fixes: #3251

Prep work for #3251

They are currently triggering on readline/libedit includes, preventing
the sanitizer job to fail for a good reason. We still have developer
warnings as part of our circle and vtest ci.

Better diff with the --word-diff option.

There's no convenient way to parameterize configure arguments to pass
them on to distcheck's configure step. We already run distcheck on
circle ci and vtest, so it's simpler to simply run the check target.

Refs 26e92d15

The actual agorithm was found with `gperf` but its output is
not directly usable and include/tbl/http_headers.h change so
infrequently that this step is not automated.  (Asserts protect
against overlooking this step if new headers are added to
the table.)

The sanity check of our defaults shouldn't be affected by sanitizer
overhead.

To get the same checks for the code we generate with libvcc.

Spotted by ubsan.

The main discussion of this topic is going to happen in #3600, but
whatever the outcome, the fini method does not happen within the task,
so we must not dereference any cached pointers (unless we own them).

Closes #3606

And doing so plug the director lock leak spotted by asan.

Refs 47586588
Refs 6e9b8d6d

Without this fix workers could busy-spin when idle under debug+vtc_mode.

Also document `table` and `log` feature flags.

Fixes #3269