Add input validation to TokenUsageMetadata constructor

[Copilot]

Addresses feedback from #2779 to validate token counts from external API responses in the TokenUsageMetadata constructor.

Changes

• Validation logic: Rejects non-finite (NaN, Infinity), negative, or unreasonably large (>10M) token counts
• Error handling: Throws descriptive errors for invalid inputs to catch API response issues early
• Documentation: Added rationale for 10M limit (sanity check allowing large context windows like Claude's 200k)
• Test coverage: 12 unit tests covering valid and invalid input scenarios
• Code style: Fixed log message formatting inconsistency in agentIntent.ts

export class TokenUsageMetadata {
	private static readonly MAX_TOKEN_COUNT = 10_000_000;

	constructor(
		readonly promptTokens: number,
		readonly outputTokens: number,
	) {
		if (!Number.isFinite(promptTokens) || !Number.isFinite(outputTokens)) {
			throw new Error('Token counts must be finite numbers.');
		}
		if (promptTokens < 0 || outputTokens < 0) {
			throw new Error('Token counts cannot be negative.');
		}
		if (promptTokens > MAX_TOKEN_COUNT || outputTokens > MAX_TOKEN_COUNT) {
			throw new Error('Token counts exceed maximum allowed value.');
		}
	}
}

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.