sim report: word an empty-side R profit factor as insights does (#1000)

[eaitbrahim]

Closes #1000

What changed

The keel sim report's edge table printed the engine's raw Profit factor (R): inf when the R sample had no losing trade, and 0.000 when it held only scratches or a zero-R win. Those cells now use the wording that insights summary, the Rules page and rules backtest already use (#986/#998):

R sample	before	after
wins only	inf	n/a (no losses yet)
scratches only	0.000	n/a (no wins or losses yet)
zero-R win, no loss	0.000	n/a (no losses yet)
losses only	0.000	0.000 (a measured value, unchanged)
no R at all	n/a	n/a (unchanged)

• insights.profit_factor_r_reason_of(stats) sits next to profit_factor_reason_of. It calls the same _profit_factor_reason on the R sample (n_trades - n_excluded_no_risk) and decides each side by counting n_wins_r / n_losses_r, as the issue asks. It never reads a zero.
• report._profit_factor_r_cell renders the cell through undefined_reading, so the wording has one source and is not copied.
• BacktestResult / profit_factor_r, the promotion gates and pool_stats are unchanged. The test checks that the engine value is still Infinity after rendering.

Ruling

Change the display rather than keep inf. The old _r_cell docstring called inf deliberate, so I checked each reason it could still hold:

• Machine-read output: nothing parses the edge-table Markdown. The only readers are tests and people, and the figures stay on BacktestResult.
• Column width or sorting: none applies to a Markdown table.
• Units: a profit factor is a ratio, so an empty side means the same thing in R as in money.

None of these holds, so the docstring is updated. inf remains only for the degenerate case where both sides have trades but the losing R sums to zero.

Tests

• test_edge_section_words_an_empty_side_profit_factor_as_insights_does: one test covering all seven rows above plus the pooled row. It also asserts the fixtures' raw engine values first, so no case passes vacuously.
• Mutation-checked 5 pins: reading a zero instead of counting (win side and loss side), printing the raw cell, dropping the no-R guard, and dropping the reason branch. Every mutant failed the suite, and each mutated source was confirmed to differ from the saved copy with cmp.
• Full pytest: 8459 passed, 3 skipped. mypy, ruff check and ruff format --check are clean.

🤖 Generated with Claude Code

[eaitbrahim]

Review of #1003 (issue #1000). The wording is right: it matches rules backtest and insights summary for every case. Sides are decided by counting, and BacktestResult and the gates are unchanged. No other surface renders profit_factor_r, and nothing parses the edge table. Three comments below: one minor (layering) and two docstring nits.