The counter cache_hit_grace counts the number of grace hits. To be
precise, it counts the number of times lookup returns an expired
object, but vcl_hit is called and decides to return(deliver).

Every time cache_hit_grace is incremented, cache_hit is also
incremented (so this commit does not change the cache_hit counter).

I'm guessing this is due to rounding. All test cases involving file
stevedore has a minimum 10m file in the test, was silly to attempt a
smaller one in this test.

Fixes: #2496

of sourcefile names over the project.

Fixes #2448

for example, mgt_shm_atexit could race in two processes, one in
VSMW_Destroy and the other in system("rm -rf")

Closes #2484

... now that I've found the layer 8 bug

Thank you to @bsdphk

Ref #2484

I had left in superfluous stuff from bug analysis.

Ref: #2319

Creating "backend synthetic" content was impossible whenever a fetch
had already started, unless storage was assigned (again) explicitly
from VCL.

Fixes #2494

The oa_present member acts as a filter at the level above the stevedore
to check set attributes efficiently.

We missed to clear it when recreating freeing an abandoned backend
fetch object in vcl_backend_error.

Fixes #2319

sanely, in this case by failing VCL initialization.

Fixes #2036

order uncertainty.

(Replaces cache_backend_cfg.c::backend_find)

We only want to return the connection early to the waiter when the
request is empty. Correct the read timeout calculation to reflect
that.

Thanks to Stackpath for helping to debug this issue.

Fixes: #2492

The documentation on directors needs an update too.

If workspace was exhausted, vmod_blob would fail yet still leave a
reservation which would likely trigger a WS_Reserve() assertion
failure in later code trying to reserve the workspace.

Fixes #2488

Thank you to @jarro2783 for the report and @dridi for the analysis.

beeaa19c had left 32 bytes on the client
workspace for the /baz case, which was sufficient to write resp.http.x-of
for 4-byte alignment, but not 8byte alignment:

* "false\0"		=  6b
* "x-of: false\0"	= 12b

Let's see if this fixes #2482