feat(runtime): add report-only intraday liquidity diagnostics

[StackOverFlow11]

Summary

P-I5f: an opt-in, report-only execution liquidity diagnostic for the
intraday tail-rebalance path. For each desired rebalance trade
(|target_weight - current_weight|) it sizes the trade against the SELECTED
execution-minute 1min bar's traded amount (RMB) at a max participation rate and
reports a capacity ratio:

desired_notional      = |target_weight - current_weight| * portfolio_notional
bar_capacity_notional = execution_minute_amount * max_participation_rate
capacity_ratio        = bar_capacity_notional / desired_notional   (>=1 covers the trade)

It uses ONLY the execution-minute bar already chosen by the tail execution model —
never a future bar, daily amount/volume/close, or any EOD proxy.

Report-only: it never changes fills, can_buy/can_sell, blocked reasons,
target weights, achieved holdings, turnover, cost, NAV, factor scores, MMP
grouping, alpha, or portfolio construction. Default-off, so every existing
config validates and behaves byte-identically.

Changed files (7)

• qt/config.py — nested LiquidityDiagnosticsCfg under intraday (enabled=false
  default; when enabled: positive portfolio_notional, max_participation_rate in
  (0,1], and only mode=report_only — each with a readable error).
• runtime/backtest/engine.py — additive, default-off record_rebalance_plan
  flag + rebalance_plan_log() capturing the exact (target, current) weights the
  engine used. Default False keeps the loop byte-identical (no plan rows, no
  behaviour change).
• runtime/backtest/event_models.py — additive accessors exposing the raw up/down
  limit blocked-key sets, so the diagnostic skips already-blocked trades without
  reclassifying their original reason.
• runtime/intraday_liquidity.py (new) — pure formula primitives + the report-only
  builder.
• qt/intraday_tail_framework.py — wires the diagnostic into the intraday-tail
  report path only when enabled; new section states it did not alter fills/NAV.
• config/phase_i5f_intraday_liquidity_diagnostics.yaml (new) — SSE50 smoke based on
  the I5b config; illustrative 10,000,000 RMB sizing.
• tests/test_i5f_intraday_liquidity.py (new, 17 tests).

Smoke (cache-only minute, real SSE50)

run-phase-i5a-intraday --config config/phase_i5f_intraday_liquidity_diagnostics.yaml

• periods=3, covered=58/58, stk_mins_live_calls=0, final_nav=0.976448.
• final_nav=0.976448 is byte-identical to the accepted I5b smoke → enabling
  the diagnostic did not move the book (report-only confirmed end-to-end).
• diagnostics: 43 desired trades, 1 feasibility-blocked (the I5b up-limit buy,
  reason kept), 0 missing-capacity, 42 inspected, 20 below 1.0x;
  capacity ratio min 0.122 / p10 0.222 / median 1.104 / p90 2.730.
• No alpha/performance/tradability claim — a single execution minute on SSE50
  large-caps often can't absorb a 1M RMB trade at 5% participation; that's the point.

Gates (all green)

• targeted tests/test_intraday_execution.py test_i5b_execution_feasibility.py test_i5d_mmp_quintile.py: pass.
• full pytest -p no:cacheprovider: 696 passed (678 baseline + 17 new I5f tests
  • 1 from the test_cache_config.py config-glob auto-covering the new config).
• ruff: clean · validate-config: 19/19 OK · run-phase0: ic_mean=0.9600,
  annual_return=0.8408 unchanged.
• git diff --check main...HEAD: clean; no merge markers.
• secret scan: real token value 0 hits in changed files / report / log / smoke
  output; report/log contain no .config.json / tushare.token.

Report-only invariants (tested)

• enabling the plan log does NOT change NAV / turnover / cost / holdings vs disabled
  (engine assert_frame_equal);
• capacity uses the execution-minute amount only (a later same-day minute's huge
  amount is ignored);
• existing missing-bar / missing-price / raw stk_limit blocks keep their original
  reasons and are never reclassified as liquidity blocks.

Non-goals honored

No partial-fill / volume-cap enforcement; no change to simulate_fills /
SimExecution / BacktestEngine / IntradayTailEventModel behaviour beyond
additive + default-off; no MMP research or I5d/I5e rerun; no daily-volume proxy; no
D6 / schema registry / cache refactor; no progress-doc edits (AGENTS.md /
CLAUDE.md / contract untouched — separate post-acceptance goal); PR #38
untouched.

⚠️ Do not merge until Codex acceptance.