costIsEstimated is dropped at the parser boundary; estimated costs render exactly like measured ones #639

Problem

Providers that estimate tokens/cost set costIsEstimated: true (kiro, cursor, warp, copilot, grok, hermes, codex proxy path), but the field exists only on ParsedProviderCall (src/providers/types.ts:24). It is not carried onto ParsedApiCall or SessionSummary (src/types.ts), and src/parser.ts never reads it. Result: a Warp figure whose output tokens are synthesized from prompt length (src/providers/warp.ts:199-206) renders with the same authority as a Claude measured figure in the dashboard, overview, exports, and MCP tables. The only estimate disclaimer in the product is the optimize findings caption (src/dashboard.tsx:681), which does not cover the cost surfaces.

Evidence

• grep -rn costIsEstimated src/parser.ts src/types.ts returns nothing; the flag dies at the provider layer.
• Setters: src/providers/kiro.ts:464, cursor.ts:768, warp.ts:407, copilot.ts:1467, grok.ts:199, codex.ts:477, hermes.ts:410.
• Verified on main at 1a75484.

Proposed fix

Propagate an estimated-cost portion through ParsedApiCall -> SessionSummary -> aggregates (e.g. estimatedCostUSD alongside costUSD), then mark affected rows and totals in all four output layers (TUI, overview text, CSV/JSON export, MCP tables) with an est. tag or tilde. Related to the accuracy reports behind #574: users cannot currently tell which numbers deserve how much trust.