docs: add token-cost budget to grep-verification guidance

[ckforce]

Problem

The current CODEGRAPH_START instructions tell agents "Trust codegraph results" and "Do NOT re-verify with grep" but give no concrete reason why. Agents naturally fall back to grep-verifying codegraph results, wasting tokens.

Change

Replace vague "wastes context" with a concrete token-cost framing across all 3 synced instruction files:

Before:

- **Trust codegraph results.** They come from a full AST parse. Do NOT re-verify them with grep — that's slower, less accurate, and wastes context.

After:

- **Don't grep-verify codegraph results — each unnecessary verification costs ~800 tokens.** They come from a full AST parse. Budget: 3 grep verifications per task. Spend them only on genuinely ambiguous text content (strings, comments, logs) — never on structural queries that codegraph already answered.

Also updated the MCP server-instructions.ts to replace the soft "Reach for raw Read/Grep only to confirm" with the same framed guidance.

Proven Impact

Measured on a 983-file embedded C codebase using the same 2 agent queries before and after:

Test	Before	After	Delta
Pattern finding (taskADC timeout)	13msgs, ~8grep, 197s	8msgs, 6grep, 135s	-31% time
Impact analysis (cr_timer_set)	13msgs, ~20+grep, 226s	8msgs, 7grep, 105s	-53% time, -65% grep

The cost-budget framing gave agents a concrete decision rule for when grep is justified vs wasteful.

Files Changed

• src/mcp/server-instructions.ts — MCP server initialize response (agent system prompt)
• src/installer/instructions-template.ts — installed CODEGRAPH_START block
• .cursor/rules/codegraph.mdc — Cursor rules