[compiler] Avoid quadratic Array.shift() in three worklist traversals #37582

Summary

Fixes #37414.

Three BFS/worklist traversals in the React Compiler dequeue with Array.prototype.shift():

• 
  compileProgram's worklist of functions to compile (Entrypoint/Program.ts)


• HIR reachability shrinking, _shrink (HIR/HIRBuilder.ts)


• control-flow post-dominator computation, postDominatorsOf (Inference/ControlDominators.ts)

Array.prototype.shift() is O(n) because it re-indexes every remaining element, so looping shift() until the queue is empty makes each of these traversals O(n²) instead of O(n) on large inputs.

All three queues are append-only worklists (items are only ever pushed, and visited/seen-set checks prevent re-processing), so the front element never needs to be physically removed. This PR replaces each shift()-based loop with an index cursor (let queueIndex = 0; while (queueIndex < queue.length) { const item = queue[queueIndex++]; ... }) that advances over the same backing array. Traversal order (FIFO) and dequeue/visit behavior are unchanged — only the O(n) removal cost per dequeue is eliminated.

How did you test this change?

• 
  yarn workspace babel-plugin-react-compiler lint — passes


• 
  tsc --noEmit for babel-plugin-react-compiler — passes


• 
  yarn workspace babel-plugin-react-compiler test (fixture/snapshot suite) — passes, no fixture output changes (traversal order is identical since these are FIFO append-only worklists)