fix(v4.0.0): three defects -- a refused device write, a failed starting state, a case that QUITs

A block write the device refuses after the kernel has taken the node's
copy: the node was told "Storage refused" while the kernel's cache kept
the new data, to be read back and perhaps written later.  The kernel now
puts its own copy back as it was.  Tested with a device whose writes can
be made to fail; it failed first.

POST's runner: a case whose starting state could not be set says so, with
what the node said, where it showed nothing; and a case that ends with
QUIT fails, as it did when the cases were a capsule, where the runner had
taken it for a completed line.  Both tests failed first.

The v4 boot says "Artemis: virtio-blk attached" only when the attach
worked.

make -C v4 test, sanitize and hosted-check pass; amd64, aarch64 and
riscv64 boot, POST 538 of 538, dict_hash 0x5f0a949a6fc8ef2b on all six:
logs/20261007-135741, -140004, -140329.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

v4/include/v4/blocks.h

@@ -34,7 +34,8 @@
  * by SAVE-BUFFERS or when its slot is wanted for another block.  (v3's
  * UPDATE copies the slot to the kernel at once; the standard is followed.)
  * A slot whose block is not marked is taken before one whose block is.  A
- * block that storage would not take is let go of: its slot is empty.
+ * block that storage would not take is let go of: its slot is empty, and
+ * the kernel keeps the block as it was before, not what was refused.
  *
  * When the kernel's chain of devices changes, everything in the slots is
  * let go and nothing of it is written, as v3 does (blk_vm_check_epoch).

v4/include/v4/post.h

@@ -36,7 +36,10 @@ extern const unsigned     v4_post_case_count;
  * case's lines one at a time, keeping what the node prints and showing none
  * of it; and judges.  A case passes when it ended in an error exactly if
  * one was expected and, if none was, its stack and its printing are what
- * the table has.  It has ended in an error if any of its lines did.  What
+ * the table has.  It has ended in an error if any of its lines did.  A
+ * case one of whose lines ends with QUIT fails: it went back to the
+ * terminal and did not complete.  So does one whose starting state could
+ * not be set, and what the node said of that is shown.  What
  * the cases define is still in the node's dictionary when the runner
  * returns: whoever runs POST puts the dictionary back afterwards, as v3
  * does after each of its suites (v3/src/test_runner/test_common.c,