fix(insights): sanitizer warning — raise threshold + add real leak signal

The previous 500-token threshold fired on virtually every authenticated
session because rrweb masks inputs at source and stripQuery() removes
URL params before sanitize() runs — so 0 matches is the normal case,
not an anomaly. Confirmed on session 019e455e: 9m32s authed session
with ~1093 tokens, 0 matches, no actual PII leak.

- Raise broad threshold 500→2000 tokens
- Add precise leak detector: any unmasked \`input [tag] "text"\` line
  in the compressed timeline AND authed AND 0 matches → alert. This
  catches the actual failure mode (PostHog source-side masking broken)
  instead of guessing from total token volume.
- Sync UI warning threshold to match (500→2000).

apps/worker/src/jobs/compress-sessions.ts

@@ -73,9 +73,19 @@ export async function runCompressSessions(): Promise<{ compressed: number; faile
         customEvents,
       );
 
-      // Sanitization sanity check: large token output but zero matches is an anomaly.
-      // Don't abort here (Phase 6a defers LLM), but flag it.
-      const anomaly = out.tokenCountEstimate > 500 && out.sanitization.matches === 0;
+      // Sanitization sanity check.
+      // Most sessions legitimately have 0 PII matches — rrweb masks inputs at
+      // source and stripQuery() removes URL params before sanitize() runs.
+      // Two anomaly modes worth alerting on:
+      //   1) very large output with 0 matches (covers all leaks broadly)
+      //   2) any unmasked input ("text") slipped past PostHog masking
+      // Mode 2 is the precise leak signal — unmasked inputs in the compressed
+      // text mean the source-side mask failed. Format is `input [tag] "text"`.
+      // Masked rows look like `input [tag] (masked)`.
+      const unmaskedInputCount = (out.timeline.match(/^\d\d:\d\d → input [^\n]*?"[^"]/gm) ?? []).length;
+      const anomaly =
+        (out.tokenCountEstimate > 2000 && out.sanitization.matches === 0) ||
+        (unmaskedInputCount > 0 && out.sanitization.matches === 0 && s.isAuthenticated);
 
       const minioKey = `sase/${dateFolder(s.startedAt)}/${s.id}.txt`;
       await putText(COMPRESSION_BUCKET, minioKey, out.timeline);
@@ -109,7 +119,9 @@ export async function runCompressSessions(): Promise<{ compressed: number; faile
       });
 
       if (anomaly) {
-        console.warn(`[compress] sanitization-anomaly session=${s.id} tokens=${out.tokenCountEstimate}`);
+        console.warn(
+          `[compress] sanitization-anomaly session=${s.id} tokens=${out.tokenCountEstimate} unmasked_inputs=${unmaskedInputCount}`,
+        );
         void alertSanitizationAnomaly({
           sessionId: s.id,
           tokens: out.tokenCountEstimate,