v0.4.0

Changed

• BREAKING: planImportMerge now returns a fourth bucket, stale: the
  existing source-imported rows (nativeId non-null) that the batch neither
  matched by nativeId nor claimed by content -- orphans whose source id
  vanished, wallet-edited ones included. It is advisory: the planner still never
  deletes anything, so a caller decides whether to offer cleanup, re-link, or
  ignore them. Wallet-only rows (null nativeId) are never stale. Callers that
  exhaustively destructure or type the return value need updating.
• planImportMerge falls back to a content match when an incoming contact's
  nativeId matches no row, so a contact whose source id churned -- notably
  Android's aggregate contact id, which changes when the OS re-links raw
  contacts across accounts -- rebinds to its existing row instead of duplicating
  it. Only rows holding a nativeId that no contact in the batch claims are
  candidates, and each row can be claimed by at most one incoming contact (first
  match wins, in incoming order). A shared DID matches on its own, even when the
  display names differ. Otherwise a candidate matches when the normalized
  display names are equal and non-empty AND the two share at least one
  normalized phone or email -- or, when both sides have no phones, emails, or
  DIDs at all, on the name alone. Phones compare by digits when present, else
  by the digits of number; emails compare trimmed and lowercased; names
  compare NFKC-normalized, lowercased, and whitespace-collapsed; DIDs are taken
  from urlAddresses via getDids and compared verbatim. A matched row that
  was never wallet-edited is overwritten, which rebinds it to the new
  nativeId; a wallet-edited one is skipped, keeping the invariant that a
  wallet edit is never overwritten by an import while still not duplicating the
  row. Contacts and rows with a null/undefined nativeId never participate.