fix(atoms): clean up MappedSignal#C when no inner signal changes state

[bowheart]

Description

I had Claude run through the events system to try to find any bugs. It didn't on the first pass, which is great, but I pressed it and it did find an edge case on the second try. Claude's summary:

Bug Found & Fixed
The bug was in MappedSignal.ts:172 — the this.C (change events) field leaked when .set() was called with a new object reference but all inner signal values were unchanged. Since no inner signals actually changed state, j() was never scheduled and this.C was never cleared. This caused:

1. Stale events sent on the next update — the leaked this.C would be used instead of the real events from the next state change
2. Real events suppressed — the !this.C guard in r() prevented forwarding genuine events from inner signals
   The fix — one line added in the finally block of MappedSignal.set():

} finally {
  schedulerPost(this.e)
  this.C = undefined  // ← prevents stale event leaks
}

This is safe because by the time schedulerPost() returns, j() has either already consumed and cleared this.C, or it never ran and the value is stale.

It looks like a real edge case to me and not even that difficult to repro. So I'm PR'ing the fix. Claude added 38 tests before finding this, so I've committed those too.