host/vitest: migrate off removed test.poolOptions for Vitest 4

[brandonpayton]

What

Migrate host/vitest.config.ts off test.poolOptions, which Vitest 4 removed
(the dev shell now installs Vitest 4.1.9). The former poolOptions.forks.maxForks
becomes the top-level maxWorkers; pool: "forks" and teardownTimeout are unchanged.
Stale comments referencing vitest 3.2.4 / a pending 3.2.5+ fix are refreshed.

Before this change every cd host && npx vitest run printed:

 DEPRECATED  `test.poolOptions` was removed in Vitest 4. All previous `poolOptions` are
 now top-level options. … https://vitest.dev/guide/migration#pool-rework

Change

-    poolOptions: {
-      forks: {
-        maxForks: process.env.CI ? 1 : 4,
-      },
-    },
+    maxWorkers: process.env.CI ? 1 : 4,

Behavior is preserved: with pool: "forks", maxWorkers caps the number of worker
forks — 1 on CI (serialized) and 4 locally (parallel), exactly as maxForks did.

Verification (vitest 4.1.9)

• No deprecation warning after the change, for both CI states — before/after logs in
  test-runs/kd-wgzh/logs/.
• Behavior preserved — Vitest's own createVitest() resolves the config to
  pool:"forks", maxWorkers:1 under CI=1 and maxWorkers:4 unset
  (test-runs/kd-wgzh/resolved-config.txt).
• Outcome lists: test-runs/kd-wgzh/outcome-lists/{passed,failed,skipped}-checks.tsv
  (8 passed / 0 failed / 1 skipped).
• Not run locally: the full cd host && npx vitest run suite — this clean checkout has
  no built wasm sysroot/kernel/release artifacts that global-setup.ts needs (dev-shell
  wasm32posix-cc + wat2wasm + a built musl sysroot). No test case exercises
  poolOptions/maxWorkers beyond config resolution (verified above); CI runs the full
  host suite in the canonical environment. See test-runs/kd-wgzh/DISPOSITION.md.

Scope

host/vitest.config.ts was the only tracked file using the removed poolOptions
(sdk and packages/registry/* vitest configs never set it).

Bead: kd-wgzh

🤖 Generated with Claude Code

[brandonpayton]

Superseded by #907. The Vitest 4 configuration fix is on main through 149fb3f98, reconstructed with the serialized host-test coverage that #907 needed; this PR was not merged verbatim. Its committed test-runs/ evidence is deliberately not carried forward.