perf: reduce nameplate aura and health work

[Nnoggie]

Summary

• reduce EllesmereUI Nameplates aura rebuild work by waiting one frame for the paired UNIT_AURA payload before falling back to a full rebuild
• reuse the already-read curHealth value for health text instead of calling UnitHealth(unit) again

Method

Profiling was done with !!!Perfy instrumentation on EllesmereUINameplates only, using repeated 30-second target dummy tests. For each pass, the live !!!Perfy.lua SavedVariables file was copied before analysis, then analyzed from the copied file. NAP screenshots were used as the coarse addon-level view, and Perfy stack totals were used to identify the concrete hot paths.

The first 30-second capture on five target dummies showed Nameplates at ~174.6ms in NAP and ~177.7ms in Perfy stack time. Most cost came from UpdateAuras (~109.5ms CPU, ~11.39MB allocations) and UpdateHealthValues (~53.6ms CPU).

The aura fix came from tracing the RefreshAuras / UNIT_AURA order. RefreshAuras could run before our UNIT_AURA handler had stashed the update payload, so the code marked an owed full rebuild. The next-frame drain then discarded the gated payload and called UpdateAuras(nil), bypassing relevance checks and causing repeated full aura scans. The fix adds _auraAwaitingUnitAura: wait one frame for UNIT_AURA; if it arrives, process the gated update, otherwise do one full fallback rebuild.

The second 30-second capture showed Nameplates down to ~111.9ms in NAP and ~99.2ms in Perfy. Health then became the next clear path. UpdateHealthValues already reads curHealth, but health text called UnitHealth(unit) again. Reusing curHealth removed that duplicate C API call.

Final observed range after both fixes was ~95-100ms Perfy stack time and ~95-98ms NAP total for Nameplates in the same 30-second target dummy setup. A speculative extra aura guard was tested and dropped because it did not improve the next capture.