fix(usage): clamp negative input token counts to zero

Some OpenAI-format providers (via pi-ai) pre-subtract cached_tokens from
prompt_tokens upstream.  When cached_tokens exceeds prompt_tokens due to
provider inconsistencies the subtraction produces a negative input value
that flows through to the TUI status bar and /usage dashboard.

Clamp rawInput to 0 in normalizeUsage() so downstream consumers never
see nonsensical negative token counts.

Closes #30765

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/agents/usage.test.ts

@@ -88,6 +88,23 @@ describe("normalizeUsage", () => {
     });
   });
 
+  it("clamps negative input to zero (pre-subtracted cached_tokens > prompt_tokens)", () => {
+    // pi-ai OpenAI-format providers subtract cached_tokens from prompt_tokens
+    // upstream.  When cached_tokens exceeds prompt_tokens the result is negative.
+    const usage = normalizeUsage({
+      input: -4900,
+      output: 200,
+      cacheRead: 5000,
+    });
+    expect(usage).toEqual({
+      input: 0,
+      output: 200,
+      cacheRead: 5000,
+      cacheWrite: undefined,
+      total: undefined,
+    });
+  });
+
   it("returns undefined when no valid fields are provided", () => {
     const usage = normalizeUsage(null);
     expect(usage).toBeUndefined();