fix(core): register MagicDocs on tool-result push, not only assistant push

[bug-ops]

Summary

• MagicDocs registration never fired for a single read-then-respond turn (the canonical --bare -p "read X" shape), or for a turn that exited the native tool loop via shutdown/user-cancel/doom-loop/max-iterations with an unpaired tool result as the last message.
• Root cause: detect_magic_docs_in_messages() only scanned when the last pushed message was Role::Assistant, deferring detection to the next assistant push. Two terminal-response sites in tier_loop.rs also bypassed push_message() via a raw self.msg.messages.push(), so the detector never ran for the turn's final response.
• Fix: both raw pushes now route through push_message(), and the detection guard also scans when the last message is a Role::User message carrying ToolResult/ToolOutput parts, so detection fires uniformly at the point content actually arrives, covering all native-loop exit paths (not just the two terminal push sites).
• Side effect: the two corrected push sites now also update cached_prompt_tokens and last_assistant_at, closing a pre-existing token-accounting gap on the semantic-cache-hit and plain-text-response paths.
• Not a --bare-conditional gate: with_bare_mode required no change. This was a general tool-loop message-push coverage gap that --bare's single-shot invocation pattern exposed deterministically; interactive multi-tool-call sessions usually masked it via a retroactive scan on the next assistant push.

Closes #6127

Test plan

• Added 3 regression tests in crates/zeph-core/src/agent/tool_execution/tests/sanitize_and_native_tests.rs, each driving the real Agent::process_response() path (not a hand-built message history): single read-then-respond turn, semantic-cache-hit branch, and loop-exit via max_iterations exhaustion with an unpaired trailing tool result. All confirmed to fail on pre-fix code and pass on the fix.
• cargo +nightly fmt --check clean
• cargo clippy --profile ci --workspace --all-targets --features "desktop,ide,server,chat,pdf,scheduler,testing" -- -D warnings clean
• cargo nextest run --config-file .github/nextest.toml --workspace --features "desktop,ide,server,chat,pdf,scheduler" --lib --bins — 13180 passed, 35 skipped, no regressions
• RUSTFLAGS="-D warnings" RUSTDOCFLAGS="--deny rustdoc::broken_intra_doc_links" cargo doc --no-deps --workspace --features "desktop,ide,server,chat,pdf,scheduler" clean
• CHANGELOG.md updated
• Testing playbook and coverage-status.md updated