fix(llm): count escalation payload_chars as characters, not bytes

[jeffrey701]

Summary

summarize_messages fills EscalationPayloadSummary.payload_chars with String::len() — the UTF-8 byte length — not the character count the field name (and the rest of the codebase) means:

payload_chars: messages.iter().map(|m| m.content.len()).sum(),

So any non-ASCII content over-reports: "café" → 5 (bytes) instead of 4 (chars); an accented household name ("José"), a non-Latin script, or an emoji device alias — all ordinary for a home assistant — inflate the count. This value isn't cosmetic: server.rs feeds summary.payload_chars into security::audit::log_provider_escalation, so the wrong number is written to the operator audit trail for "data left the device via the gated cloud path" (#570).

The codebase has a clear, tested convention that *_chars metrics are Unicode char counts — tools/dispatch/mod.rs uses .chars().count(), and telegram.rs carries a regression test literally named voice_reply_gate_char_count_uses_unicode_chars_not_bytes. escalation.rs was the lone *_chars site still using .len(). This aligns it; the existing summarize_messages_counts_chars_not_content test only used ASCII (where bytes == chars), so the bug was latent.

Closes #851

Real Behavior Proof

• I have built and run the affected code locally (or noted why I could not).
• I have verified the change end-to-end on Jetson hardware.
• I have NOT verified on Jetson hardware, and I explain the equivalent verification path or validation gap below.

Tested profile / hardware (check all that apply):

• jetson
• raspberry_pi
• portable_sbc
• laptop
• mac
• CI-only / docs-only
• Not run locally

genie-core is a Linux-only crate, so it doesn't compile on my Windows dev box — CI runs the real cargo test on Linux, exercising the added summarize_counts_unicode_chars_not_bytes test. Locally I ran the old .len() vs new .chars().count() counting side by side in a rustc harness.

What I ran

rustc --edition 2021 pr_chars_test.rs && ./pr_chars_test

What I observed

OK: OLD counted bytes (café=5); NEW counts chars (café=4); ASCII unchanged (=6).

The harness asserts OLD payload_chars("café") == 5 (byte length — the bug), NEW == 4 (char count), the same for "José" (5→4) and an emoji alias (9→6), and that ASCII content is unchanged ("abcd"+"ef" == 6, matching the existing test).

Summary by CodeRabbit

• Bug Fixes
  • Corrected escalation audit summaries to count Unicode characters accurately instead of counting encoded bytes.
  • Added coverage for messages containing accented and other multi-byte characters.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 6fd4bc76-0e6f-43d1-a650-9a380f43f5fd

📥 Commits

Reviewing files that changed from the base of the PR and between 7c67906 and cdd3224.

📒 Files selected for processing (1)

• crates/genie-core/src/llm/escalation.rs

---

📝 Walkthrough

Walkthrough

The escalation payload summary now counts Unicode characters instead of UTF-8 bytes, and a unit test verifies that multi-byte text such as "café" is counted correctly.

Changes

Unicode payload counting

Layer / File(s)	Summary
Payload summary and validation crates/genie-core/src/llm/escalation.rs	payload_chars now sums content.chars().count(), with a test confirming "café" counts as four characters.

Estimated code review effort: 2 (Simple) | ~10 minutes

Suggested reviewers: kiannidev

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 66.67% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and accurately summarizes the main change to count escalation payload characters instead of bytes.
Linked Issues check	✅ Passed	The PR implements #851 by switching to Unicode character counts and adding regression coverage for non-ASCII content.
Out of Scope Changes check	✅ Passed	The changes stay focused on the payload character-count fix and its test coverage, with no unrelated edits.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[matedev01]

LGTM — payload_chars used String::len() (UTF-8 byte length) instead of char count, so non-ASCII content over-reported in the escalation audit trail ('café' -> 5 instead of 4). Small, correct fix to the audit-log accuracy from #648/#796. Verified: clippy -D warnings clean, escalation tests (4) pass, fmt clean.