[Devourer] Tune Void Metamorphosis fury drain constants

[taherbert]

Measured the in-meta fury drain constants in fury_state_t::fury_drain_per_second against WCL. The drain logs each 2 fury as a resource-change event under spell 473671, so both the instantaneous rate (2 divided by the time between ticks) and the full cumulative drain-event schedule are directly observable. Pulled ~100 current Devourer raid logs, 210 metas (Soul Glutton excluded).

Three values were off:

• First second after entering Void Metamorphosis drains 10/s, not 15 (the drain_stacks < 1 floor).
• Reduced drain while casting Collapsing Star or channeling Void Ray is 0.127x, not 0.15x.
• The curve tail is slightly steeper than 15 + 1.455 * exp(0.075 * s). With only the two changes above, replaying SimC's drain scheduler through each logged meta (against that meta's actual Collapsing Star / Void Ray windows) tracks in-game cumulative drain counts mid-meta but falls ~9 drain events (~18 fury) behind by meta end, so sim metas would run ~0.3s long. Refit against the cumulative schedules: 15 + 1.40 * exp(0.0775 * s) matches in-game cumulative drain events within +-2 at every checkpoint including meta end.

For reference, the old constants (15/s first second, 0.15x reduced, 1.455/0.075 curve) do not match the in-game timeline either: they run 4-9 drain events ahead through mid-meta and only land near the correct end time because the too-shallow tail cancels the too-hot early drain.

Structure checks from the same logs: the acceleration steps once per second exactly like drain_stacks (the within-second rate pattern is flat, not continuous-time), the quantum is exactly -2 per event, and Soul Glutton's 1.333x multiplier matches. Coverage past ~52s in meta is thin (few logged metas survive that long), so the tail beyond that is the fit's extrapolation.