fix(examples, record): follow the farms that answer for their own members

`compat:vulkan-runtime` published a farm whose members needed four sonames
nothing on the artifact's search path carried. The three examples that pin it
moved to 2026.09.10, where the farm closes over its own members and reaches the
host only through a named `*-host-link` package.

Measured on this host, the vulkan example's own record before and after:

    members 55, walked 55, missing 4, dangling 0
    members 60, walked 59, missing 0, dangling 1

The one remaining finding is `libcrypto.so.1.1`, and it is dangling rather than
missing because the publishing package now says so: OpenSSL 1.1 is end-of-life
upstream, the ecosystem publishes 3.x, and only NVIDIA's PKCS#11 provider asks
for it. `libcrypto.so.3` comes from `xim:openssl` and `libnvcuvid.so.1` from
`xim:nvidia-video-host-link`; no branch harvests a new file from /usr/lib.

The example still runs on the device: `12 24 36 48` and
`device: NVIDIA GeForce RTX 4080`.

Section 7 of the design record no longer files those four findings as open. It
records what repairing them found instead: the defect was in three farms rather
than one, mcpplibs/mcpp-index#376 named the wrong one -- the four came from
`compat:opencl-runtime` -- and the reason written against the fix had measured
the set before the driver's settings GUI was excluded.

Follows mcpplibs/mcpp-index#380 and openxlings/xim-pkgindex#801.

Author: speak-agent

.agents/docs/2026-09-09-dlopen-surface-and-two-unwinders.md

@@ -603,11 +603,28 @@ already filed as mcpp-index#376, reached this time through
   | `libnvidia-pkcs11.so.550.144.03` | `libcrypto.so.1.1` | same |
 
   All four are real and all four are the same shape as mcpp#596 -- a farm that
-  mirrors a driver family and stops one library short. They are NOT repaired
-  here: a check whose first catch is its author's own package should report it,
-  not quietly absorb it, and the repair belongs to `compat:vulkan-runtime` with
-  its own criterion, filed as mcpplibs/mcpp-index#376. On a runner with no NVIDIA driver the farm is nearly empty
-  and the check is silent, so this does not appear in CI.
+  mirrors a driver family and stops one library short. They were filed rather
+  than absorbed, as mcpplibs/mcpp-index#376, and repaired there in
+  mcpplibs/mcpp-index#380. On a runner with no NVIDIA driver the farm is nearly
+  empty and the check is silent, so this never appeared in CI.
+
+  WHAT THE REPAIR FOUND THAT THE FILING DID NOT. The defect is in three farms,
+  not one, and the issue names the wrong one: the four findings above were
+  published by `compat:opencl-runtime`, while `compat:vulkan-runtime` had five
+  and `compat:glx-runtime` -- which had no closure pass at all -- had thirty.
+  The cause is one rule: membership is decided by a filename pattern while
+  completeness was decided against the ICD manifests, so the half of each farm
+  the pattern exists for is the half nothing verified. The reason recorded
+  against seeding the closure from the farm turned out to have measured the
+  set BEFORE `never_farm_patterns` removed the driver's settings GUI; measured
+  after, it adds five libraries and no GUI stack.
+
+  The repair also holds the host surface down rather than widening it. Every
+  soname a member needs now has a written answer -- an ecosystem package
+  declared in `deps`, a `*-host-link` sentinel for what cannot be
+  redistributed (openxlings/xim-pkgindex#801 adds one for the codec
+  user-space), a named entry in `UNSERVED` with its reason, or a warning at
+  install time. No branch harvests a new file from `/usr/lib`.
 
 ## 8. Order
 

examples/09-heterogeneous/multi-backend/mcpp.toml

@@ -104,7 +104,7 @@ cuda-driver = "2026.09.05"
 # is a payload and the hardware ones are the host's.
 [target.'cfg(accelerator = "vulkan")'.dependencies.compat]
 vulkan         = "1.4.357.0"
-vulkan-runtime = "2026.09.07"
+vulkan-runtime = "2026.09.10"
 
 [build]
 # The device sources carry the accel they are for. A CONSTRAINED glob gates

examples/09-heterogeneous/vulkan/app/mcpp.toml

@@ -22,7 +22,7 @@ plugins = { version = "0.5.2", features = ["rules-spirv"], host-module = true }
 # below is a payload and the hardware ones are the host's.
 [dependencies.compat]
 vulkan         = "1.4.357.0"
-vulkan-runtime = "2026.09.07"
+vulkan-runtime = "2026.09.10"
 
 # ONE payload here, and the shader compiler is not it: `mcpp.rules.spirv`
 # declares `xim:glslang` for itself. What stays is a DEVICE, and the boundary is

examples/10-graphics/offscreen/mcpp.toml

@@ -35,7 +35,7 @@ vulkan = "1.4.357.0"
 # concern by construction: macOS resolves through dyld and Windows through the
 # PE loader, and neither has a private loader to work around.
 [target.'cfg(linux)'.dependencies.compat]
-vulkan-runtime = "2026.09.07"
+vulkan-runtime = "2026.09.10"
 
 # A Vulkan device that needs no GPU, so this example runs on a machine that has
 # none. It is a DEVICE and therefore a payload; the drivers a real GPU needs are