Skip redundant toProxy() calls in walkSync by caching proxy across enter/exit phases

[alanzabihi]

Hypothesis

In walkSync (lazy-result.js:525-574), node.toProxy() may be called up to 4 times for the same node: enter listeners (line 543), keyed enter (550), exit listeners (563), keyed exit (570). While toProxy() caches the Proxy on the node, each call still involves: function call overhead, WeakMap lookup (proxyCache), and conditional check.

The code already partially optimizes this with the local proxy variable, but it still calls toProxy() at each phase if the previous phase had no visitors. Cache the proxy unconditionally at the start of walkSync for any node that has ANY listeners:

// Pre-check: does this node type have any listeners at all?
let hasAnyVisitor = listeners[type] || listeners[type + 'Exit']
if (key) hasAnyVisitor = hasAnyVisitor || listeners[type + '-' + key] || listeners[type + 'Exit-' + key]
let proxy = hasAnyVisitor ? node.toProxy() : null

This reduces toProxy() calls from up to 4 per node to exactly 0 or 1.

Editable surface

• lib/lazy-result.js — restructure walkSync to call toProxy at most once per node

What's different from prior work

• Skip proxy wrapping for leaf nodes in visitSync and walkSync #40 tried lightweight leaf proxies (neutral — toProxy already caches). This thesis doesnt change the proxy itself, it reduces the number of toProxy() CALLS (function + WeakMap overhead).
• Skip proxy wrapping in walkSync for nodes without exit listeners #36 tried skipping proxy for nodes without exit listeners. This is more precise: skip only the redundant calls, keep the proxy when needed.
• walkSync already has a local proxy variable but conditionally initializes it multiple times.

Expected impact

METRIC_B improvement of 1-3ms. Combined with other optimizations in the same thesis, could contribute to crossing tolerance.

[alanzabihi]

Polyresearch approval: thesis #67.