Verify whether LR2 counts empty POORs in the result POOR tally

[nulltask]

Context

From the BMS spec-compliance audit and the wave-3 LR2 research. The current implementation does not count empty POORs (空POOR) into summary.poor — that slot is reserved for missed POORs — and does not break combo (documented in docs/player-spec.md, "Blank keystroke" section).

Open question

Whether real LR2 adds empty POORs to the result-screen POOR count (and BP):

• The wave-3 research found this to be the prevailing community understanding, with circumstantial evidence (LR2 shows no separate empty-POOR stat; beatoraja-migration writeups list "can see empty-POOR separately" as a beatoraja advantage), but no explicit first-party source.
• lr2oraja's LR2-compat judge property confirms the gauge side (-2 via the miss slot) and the no-combo-break behavior, but not the result-tally behavior.

Action

Verify against real LR2 (a recorded result screen comparison with counted key presses would settle it). If LR2 does count them:

• add empty POORs to the result POOR tally in packages/player/src/core/engine.ts (the EMPTY_POOR branch currently skips applyJudgeToSummary),
• keep combo / EX-score untouched,
• update docs/player-spec.md / .ja.md and the engine tests that pin the current behavior.

[nulltask]

Resolved by first-party LR2 evidence: empty POORs ARE counted

Reverse-engineering the actual LR2 (LR2body.exe, beta3) settled this. LR2's own changelog history.txt, entry dated 080317:

空POORが計上されないバグを修正しました。
最小B+Pが記録されないバグを修正しました。

Translation: "Fixed a bug where empty POORs were not counted. Fixed a bug where the minimum B+P was not recorded."

So real LR2 does count empty POORs into the POOR / B+P statistics (this was a fixed bug — the final behavior is "counted"). Our implementation deliberately excludes empty POORs from summary.poor, which is a confirmed divergence.

Action

• In packages/player/src/core/engine.ts, the EMPTY_POOR branch should increment the POOR tally (and feed B+P), while keeping combo intact and EX-score unchanged (combo non-break is still correct — lr2oraja's combo[emptyPoor] = true).
• Update docs/player-spec.md / .ja.md ("Blank keystroke" section currently states empty POORs are NOT counted) and the engine tests that pin the current no-count behavior.

This is a behavior fix, not just a question — relabeling as a bug.