Aggregate cost renders $0.00 for positive sub-micro spend

What happens

costevent.Event.Micros() = math.Round(CostUSD * 1e6) in costevent.go rounds sub-half-micro costs to 0. The event is still marked priced. After aggregation, CostMicros can be 0 while PricedRequests > 0. renderCostSummary then printsCOST $0.00 — the same output as a genuinely-free priced call.

This can occur for models like Anthropic Haiku's ~$0.25/M input rate, where a 1-token priced call is $2.5e-7 and rounds to 0 micros.

Behavior impact

Small-but-real spend reads as free in the usage-graph total. The per-event cell is unaffected (it renders through formatUSDCell's <$0.0001 floor against the raw CostUSD, not against the rounded micros).

Proposed fix

Add a PositiveCostRequests counter that sources from the pre-rounding CostUSD > 0 predicate:

• costevent.eventCost — new predicate exposed
• usage.Counts + Counts.Add — new field, folded at every accumulation site
• renderCostSummary — render <$0.01 when CostMicros == 0 && PositiveCostRequests > 0

Do not clamp Micros() to 1 — it changes the documented rounding semantics and lies about the underlying figure.

Ref #1000 (addressed by #1042)

Assisted-By: Claude (Anthropic AI) noreply@anthropic.com

Start with authbridge/authlib/costevent/costevent.go and trace usage.Counts, Counts.Add, and renderCostSummary through every accumulation site. Confirm the pre-rounding positive-cost count reaches the summary, then run the existing Go tests covering cost events, usage aggregation, and rendering; done means positive sub-micro spend renders < $0.01 while genuinely free priced calls still render $0.00.