Previously, we would get PROXY vxids without a Proxy line like:

    262253 Begin          c sess 0 PROXY
    262253 SessOpen       c w.x.y.z 1234 a0 a.b.c.d 85 1516787007.682753 32
    262253 Link           c req 262254 rxreq

Otherwise we might end up missing the signal and fail later when
reset the process.  All tests are passing in OSX now.

Merges #2545

I got this wrong in 69695c23 for the
case that the segment spans two pages.

Ref: #2541

mmaps from b25.vtc varnishstat:

before this change:

6598  open("/tmp/vtc.6492.2b3b5875/v1//_.vsm_child/_.VSC_cluster.322833fd50de0860", O_RDONLY) = 21
6598  mmap(NULL, 4096, PROT_READ, MAP_SHARED, 21, 0) = 0x7fb92529e000
...
6598  open("/tmp/vtc.6492.2b3b5875/v1//_.vsm_child/_.VSC_cluster.322833fd50de0860", O_RDONLY) = 21
6598  mmap(NULL, 8192, PROT_READ, MAP_SHARED, 21, 0) = 0x7fb925271000
...
6598  open("/tmp/vtc.6492.2b3b5875/v1//_.vsm_child/_.VSC_cluster.322833fd50de0860", O_RDONLY) = 21
6598  mmap(NULL, 12288, PROT_READ, MAP_SHARED, 21, 0x1000) = 0x7fb925222000
...
6598  open("/tmp/vtc.6492.2b3b5875/v1//_.vsm_child/_.VSC_cluster.322833fd50de0860", O_RDONLY) = 21
6598  mmap(NULL, 12288, PROT_READ, MAP_SHARED, 21, 0x2000) = 0x7fb92521f000

after this change:

7749  open("/tmp/vtc.7648.2c3869d9/v1//_.vsm_child/_.VSC_cluster.09720dc34dd3559c", O_RDONLY) = 21
7749  mmap(NULL, 4096, PROT_READ, MAP_SHARED, 21, 0) = 0x7f0938a28000
...
7749  open("/tmp/vtc.7648.2c3869d9/v1//_.vsm_child/_.VSC_cluster.09720dc34dd3559c", O_RDONLY) = 21
7749  mmap(NULL, 4096, PROT_READ, MAP_SHARED, 21, 0x1000) = 0x7f09389f9000
...
7749  open("/tmp/vtc.7648.2c3869d9/v1//_.vsm_child/_.VSC_cluster.09720dc34dd3559c", O_RDONLY) = 21
7749  mmap(NULL, 8192, PROT_READ, MAP_SHARED, 21, 0x1000) = 0x7f09389d2000
...
7749  open("/tmp/vtc.7648.2c3869d9/v1//_.vsm_child/_.VSC_cluster.09720dc34dd3559c", O_RDONLY) = 21
7749  mmap(NULL, 4096, PROT_READ, MAP_SHARED, 21, 0x2000) = 0x7f09389d1000

Notice that we keep the mappings at one or two pages now and do not increase
the length with the offset.

Fixes #2546

This commit partially reverts and reworks b9f7170b. We now do the "wait
for active streams" handling via the loop in h2_new_session, to ensure
the workspace is in a predictable state once HTC_RxStuff is called
again.

Fixes: #2539

No need to waste more precious human life time with stupid editing

so tell make in lib/libvcc how to invoke it.

Exposed by 062234c1: Now code changes
(as when switching branches) may require a rebuild in lib/libvcc only

... for a simple make clean, in particular not vmod_abi.h.

Otherwise, building from a trunk dist tarball will give a NOGIT
VMOD_ABI_Version after a "make clean", while the correct version
is in place for a simple make after unpacking the tarball.

In addition, instead of always initializing a transaction's storage
backend using the round-robin selection, pick Transient when we know
in advance that beresp will be uncacheable.

Closes #2533