fix(interface): clarify cached vs new input token breakdown

[not-knope]

Closes #658

What changed

• When cached input tokens are present, the stats panel now shows a three-part breakdown: Input (total), Cached (in total), and New input (uncached portion).
• When there is no cache activity, the final report keeps the simpler Input Tokens label (no redundant cache row).
• In live TUI mode with cached_tokens = 0, the cache row is still shown (Cached (in total) 0) so the sidebar metric remains visible during an active scan — same intent as before, with clearer labeling when cache is non-zero.
• Added tests/test_interface_stats.py with three regression tests covering final stats, live-with-zero-cache, and live-with-cache breakdown.

Why

Users could not tell whether cached tokens were included in the input total or additive (e.g. 20M input + 19M cached = 39M vs 20M with 19M cached). The new labels make the subset relationship explicit and surface the uncached count when caching is in use.

How I tested it

python -m pytest tests/test_interface_stats.py -v

All 3 new tests pass.

I also ran the broader test suite locally. Pre-existing failures in test_hooks.py and test_runner_rate_limit.py are due to missing pytest-asyncio in the local environment (async tests are not collected/run correctly), not related to this change.

Note

This PR was prepared with AI coding assistance (Cursor). All changes were reviewed, tested, and validated by the author before submission.

[greptile-apps]

Greptile Summary

This PR clarifies how cached input tokens are shown in usage stats.

• Adds a total, cached, and new-input breakdown when cached tokens are present.
• Keeps the simpler final-report input label when no cached tokens are present.
• Keeps the live cache metric visible at zero during active scans.
• Adds focused tests for the new stats rendering cases.

Confidence Score: 5/5

This looks safe to merge.

• No blocking issues found in the changed code.

Important Files Changed

Filename	Overview
strix/interface/utils.py	Updates LLM usage stats rendering to show cached-token breakdowns without changing the underlying usage collection path.
tests/test_interface_stats.py	Adds tests for final no-cache output, live zero-cache output, and cached-token breakdown output.

_{Reviews (1): Last reviewed commit: "fix(interface): clarify cached vs new in..." | Re-trigger Greptile}