fix(cline-cli): decline the rollup when per-message calls were all deduped

[iamtoruk]

Closes #894. Follow-up from the adversarial review of #891.

The bug

The rollup fallback in cline-cli was gated on emitted > 0, but emitted counted only calls that were actually yielded, and the seenKeys dedup check runs before the increment. So a session whose per-message calls were all suppressed by the shared dedup (a duplicated session directory reusing a session_id) kept emitted at 0 and fell through to the metadata.usage rollup, emitting the session's cost a second time under cline-cli:<id>:rollup. Over-reports; needs an abnormal on-disk layout.

The fix

Gate the rollup on whether the session carried any per-message metrics at all, tracked by a hadMetrics flag set before the dedup check rather than on the post-dedup emitted counter. A session whose messages were all deduped now correctly declines the rollup.

Verification

• New test: two directories sharing an internal session_id and message ids, the second also carrying a metadata.usage rollup, collected through one shared seenKeys set. The copy is deduped and its rollup declined, so exactly one call survives and cost stays $0.01 rather than doubling to $0.02.
• Mutation-checked: moving the hadMetrics assignment after the dedup check reproduces the original double-count (the test then sees 2 calls).
• tsc clean; all 34 cline-cli tests pass; full suite green apart from the two standing parser.test.ts failures that reproduce on main.