fix(tracing): flush log_guard on SIGTERM alongside chrome_guard

[bug-ops]

Summary

• log_guard (the rolling-file WorkerGuard in src/tracing_init.rs) had the same SIGTERM-flush gap chrome_guard had before fix(tracing): flush local trace on SIGTERM and fix span-duration fragmentation #6692 fixed it: a SIGTERM without a handler skips Drop entirely, discarding whatever log lines were still buffered in the async writer's channel. Reproduced 6/6 with a tight self-kill race, losing 127-939 lines per run; a 50ms-headstart contrast case lost 0/3, explaining why fix(tracing): flush local trace on SIGTERM and fix span-duration fragmentation #6692's informal single-run test didn't repro this.
• Generalizes fix(tracing): flush local trace on SIGTERM and fix span-duration fragmentation #6692's take-once-cell + SIGTERM-listener pattern into a new, unconditional LogGuardCell (file logging isn't gated behind the profiling feature, unlike Chrome tracing). Both TracingGuards::drop and spawn_sigterm_flush_task now take-and-flush whichever cell(s) are still Some.
• owns_sigterm_elsewhere is now unconditional (previously profiling-gated), since the log-guard flush task needs the same "don't race a graceful shutdown path" guard regardless of feature flags. Extracted the gate decision into a pure should_install_sigterm_flush_task, unit-tested against its full truth table, so a mis-wire no longer silently regresses in default builds.
• CHANGELOG documents this as a behavior change for effectively all builds/invocations (SIGTERM's default disposition is now overridden much more broadly than before), plus a bounded ~1.1s worst-case added SIGTERM-to-exit latency.
• Corrected a stale doc claim that Ctrl-C behavior was "non-truncating" and unaffected — early_ctrlc still calls process::exit(130) and bypasses guard flushing; this is called out as a known, separate limitation (not fixed here).

Closes #6693

Test plan

• cargo +nightly fmt --check
• cargo clippy --profile ci --workspace --all-targets --features "desktop,ide,server,chat,pdf,scheduler,testing" -- -D warnings
• cargo nextest run --config-file .github/nextest.toml --workspace --features "desktop,ide,server,chat,pdf,scheduler" --lib --bins (15258 passed)
• RUSTFLAGS="-D warnings" RUSTDOCFLAGS="--deny rustdoc::broken_intra_doc_links" cargo doc --no-deps --workspace --features "desktop,ide,server,chat,pdf,scheduler"
• gitleaks protect --staged
• New unit tests: tracing_guards_drop_flushes_log_guard_cell, log_guard_cell_take_is_idempotent_under_concurrent_access (genuine OS-thread race), spawn_sigterm_flush_task_does_not_touch_log_guard_before_signal, should_install_sigterm_flush_task_truth_table
• Debugger reproduced the original gap live (pkill -TERM under active log writes) before the fix, confirming the fix addresses a real, not theoretical, defect