fix(portal): make S1 fast+deterministic — pin a small chunk instead of streaming a 500k-block chunk

S1 flaked on CI with a 30s timeout. Root cause: the test left PORTAL_CHUNK_BLOCKS
at its 500k default, and the stub server streams one block per request — so
draining the first serviced chunk took ~45k real localhost round-trips (~7s idle,
ballooning past the timeout under CI load). The test's own comment ("one block at
its start") shows small chunks were the intent.

Pin PORTAL_CHUNK_BLOCKS=100 for S1: the same read-ahead + eviction accounting is
exercised in a handful of requests. S1 goes from ~7-19s (30s timeout under load)
to ~320ms, stable across repeated runs under multi-core CPU load.

Also reword the testTimeout comment to match reality (the 30s is general headroom
for the real-backoff head-probe tests + coverage instrumentation, not portal-shell).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: mo4islona

portal/portal-shell.test.ts

@@ -668,6 +668,11 @@ test('S1: rows return to baseline after chunks settle and are evicted (no double
   // fetches settle: the gate's row count must equal exactly the LIVE cache's rows — here 0, because the
   // final far interval has no data and everything behind was evicted + freed via per-fetch tokens.
   process.env.PORTAL_READAHEAD = '2';
+  // Pin a SMALL chunk (default is 500k). The stub server streams one block per request, so a 500k-block
+  // chunk means ~45k real localhost round-trips to drain the first chunk — ~7s idle, and it blows past
+  // the test timeout on a loaded CI box (the S1 flake). A 100-block chunk exercises the same read-ahead
+  // + eviction accounting in a handful of requests. (PORTAL_CHUNK_FIXED is already set in beforeEach.)
+  process.env.PORTAL_CHUNK_BLOCKS = '100';
   const srv = http.createServer((req, res) => {
     let body = '';
     req.on('data', (c) => {
@@ -727,5 +732,6 @@ test('S1: rows return to baseline after chunks settle and are evicted (no double
   } finally {
     srv.close();
     delete process.env.PORTAL_READAHEAD;
+    delete process.env.PORTAL_CHUNK_BLOCKS;
   }
 });

portal/vite.portal.config.ts

@@ -14,8 +14,8 @@ export default defineConfig({
       'src/sync-historical/portal*.test.ts',
       'src/sync-historical/realtime*.test.ts',
     ],
-    // 30s (not 15s): portal-shell's heavy fake-timer/retry tests run ~15s under coverage
-    // instrumentation, which flakes on a loaded CI server. Give per-test headroom.
+    // 30s (up from 15s): headroom for the few tests that exercise REAL backoff (head-probe retry) and
+    // coverage instrumentation on a loaded CI box. The old 15s was tight enough to flake there.
     testTimeout: 30000,
     // Coverage is scoped to the Portal layer (the whole diff vs upstream ponder). Inert on a plain
     // `--test` run: it only activates under `--coverage` (see scripts/sync-upstream.sh --coverage),