chore(fidelity): close Phase 9's loose ends — alias, stale docstrings, telemetry overflow guard

[eaitbrahim]

What & why

Closes Phase 9's loose ends, from the independent whole-phase review (all findings non-blocking; this PR is bookkeeping and robustness, not behavior):

• cli._SIM_SLIPPAGE_PCT is now an alias of backtest.SLIPPAGE_FLOOR_PCT, not a repeated 0.0005 literal. The simulate report asserts its flat-priced dollar sections cost "the flat SLIPPAGE_FLOOR_PCT per leg" — that claim was true by numeric coincidence; a retuned floor would have made the report's own cost statement silently false. Now structurally true, pinned by a test at the site that renders the claim. (portfolio_sim and paper still carry their own literals — folding them is bundled into Opt the promotion gate into per-product slippage — prerequisite-gated on a thin-asset candidate or a load-bearing restatement #335, the promotion-gate opt-in, so all remaining flat-rate sites move together.)
• Two stale fix(executor): warn loudly when a conditional entry price is overridden by market routing (#260) #332-era docstrings refreshed: _log_intent_divergence no longer claims the per-asset liquidity model "does not exist yet (slippage is one global 5bp constant applied from BTC to TON #259)" — it exists on the research side; the live path just has no statistic at fill time. ENTRY_OVERRIDE_WARN_BP's rationale now cites SLIPPAGE_FLOOR_PCT (not the CLI literal) and states honestly that "10x the slippage assumption" holds at the liquid end and narrows toward slippage is one global 5bp constant applied from BTC to TON #259's 50bp cap — plus an explicit note that the threshold's 50bp and slippage is one global 5bp constant applied from BTC to TON #259's cap 50bp are different, uncoupled constants.
• The entry-override warning's arithmetic moved inside a try: is_finite() admits extreme exponents (a rule bug like 1E+999999999 parses, is finite, compares fine), and the division raises ArithmeticError — which telemetry must swallow, matching the sibling intent_divergence's inside-the-try arithmetic. Regression-tested.
• The promotion-gate opt-in deferral now has its tracking issue (Opt the promotion gate into per-product slippage — prerequisite-gated on a thin-asset candidate or a load-bearing restatement #335), prerequisite-gated like Route a rule's conditional entry as a genuine resting order (limit/stop) — deferred until a price-conditional rule earns it #333 — the phase's one untracked decision is tracked.

Tests-first evidence

The overflow guard was written failing-first conceptually and verified directly (Decimal('1E+999999999') against a quoted preview raises through the old code, returns silently on the new); the regression case is now in test_a_preview_without_a_book_quote_is_silent_not_fatal. The alias pin lives in test_simulate_reports_per_product_slippage_beside_the_results, where the claim it guards is rendered.

Gates

• uv run ruff check keel tests packages — All checks passed!
• uv run mypy — Success: no issues found in 237 source files
• uv run pytest -q — 2887 passed, 1 skipped

Scope check

• No rail, rule, or classification touched. No behavior change except one: the entry-override telemetry can no longer raise on pathological exponents (strictly toward "never raises").