N3/P8: scale the enemy's action economy to the party

A party of N used to face one enemy swing a round against a single seat, so
each member absorbed ~1/N² of the solo incoming and cleared 100% of every T5
cell. The enemy now takes enemyActionsThisRound() attack-actions, each
re-targeted at a fresh standing seat, in both combat engines:

  - auto-resolve (simulatePartyRound): enemyRoundSwings loops the actions,
    re-rolling each target's pet procs.
  - turn engine (stepEnemyTurn): enemyAttackAction resolves one full SRD
    multiattack per action; step() no longer blanket-stamps the enemy turn to
    one seat, since a party's enemy now hits several.

The action count is a fractional expectation realised as a per-round coin-flip
(2.4 for a duo, 2N-1 for N>=3), because the integer lever is too coarse at small
N -- against a duo, 2 actions is a ~91% faceroll and 3 a ~45% grinder, with
nothing between. A light party-only enemy HP scalar (x1.15) trims the martial
ceiling that action count alone leaves at 100%.

Solo is exempt on both levers (1 action, no RNG draw; x1.0 HP), so
TestCombatCharacterization is byte-identical and the d8prereq corpus still
compares. Sim band (party of 3, T5, HP scaled): fighter 70->90%, cleric-led
27->72% -- monotonic by party size, no composition worse than soloing.

Claude-Session: https://claude.ai/code/session_017mEwUmmS7aQTP2NQXj6rUa

internal/plugin/combat_session_build.go

@@ -176,6 +176,12 @@ func (p *AdventurePlugin) partyCombatantsForSession(sess *CombatSession) ([]*Com
 			// rebuilds the identical stat block, so seat 0's copy is the fight's.
 			// Only the *player* half of the build varies by seat.
 			enemy = e
+			// Party-only enemy HP bump, re-derived each turn from the template so
+			// it never compounds. Matches the scalar startPartyCombatSession used
+			// for the initial persist; solo (roster 1) scales by 1.0.
+			if sess.IsParty() {
+				enemy.Stats.MaxHP = scaledEnemyMaxHP(enemy.Stats.MaxHP, sess.RosterSize())
+			}
 		}
 	}
 	return players, &enemy, nil