wk 0.9.0

Breaking changes

• The common well-known binary representation of POINT EMPTY (i.e.,
  POINT (nan nan)) is now handled as POINT EMPTY allowing empty points
  to roundtrip through wkb() vectors (#196, #204).
• xy(NA, NA) is now read as a null feature instead of POINT EMPTY. This
  preserves the invariant that null features can also be identified using
  is.na() (#205).
• xy(NaN, NaN) is now read as POINT EMPTY and is.na(xy(NaN, NaN))
  now returns FALSE. This means that both EMPTY and null points can roundtrip
  through xy() (#205).

Bugfixes and improvements

• wk_meta() now contains a new column is_empty, which is TRUE
  for any feature that contains at least one non-empty coordinate. This allows
  more efficient detection of features with zero coordinates (#197, #199).
• Updated PROJ data to use the latest pull of the database packaged with
  PROJ 9.3.0 (#201).
• The wk package now compiles once again on gcc 4.8 (#203, #206).
• Fixed sfc_writer() to correctly attach the classes attribute to
  sfc output with mixed geometry types (#195).
• Function sfc_writer() now has an argument promote_multi to write any
  input as the MULTI variant. This makes it more likely that an input vector
  will be read as a single geometry type (#198).
• The wk_collection_filter() now correctly increments the part_id when
  calling the child handler (@brownag, #194).