fix(cli): share one deadline across telemetry shutdown pipelines

Context

kimi web can own more than one telemetry pipeline: host-level telemetry and kap-server engine telemetry. PR #2230 intentionally did not change the CLI-wide shutdown contract.

Problem

Independent sequential timeouts allow total shutdown latency to grow with the number of telemetry pipelines. Print/web entrypoints can also drift if each implements its own timer and cleanup ordering.

Scope

• Establish one absolute telemetry shutdown deadline per CLI process exit.
• Pass the remaining budget to each owned telemetry pipeline instead of giving each a fresh full timeout.
• Keep telemetry cleanup best-effort and preserve the existing process exit result.
• Share policy without coupling CLI entrypoints to appender internals.

Acceptance criteria

• Total telemetry shutdown time stays within one configured budget across all pipelines.
• A slow or failed pipeline does not prevent later cleanup or change the command's exit result.
• Web and print lifecycle tests cover multiple pipelines, expired budgets, and rejection paths.

Follow-up to #2230.

Start with the kimi web and print entrypoints and their lifecycle tests, then read PR #2230 for the existing shutdown contract. Confirm the tests cover multiple pipelines, expired budgets, rejection paths, later cleanup, and preservation of the command's exit result within one configured budget.