Array methods

All arrays (should) have the following methods:

• constructor via the direct and most complete internal representation
• classmethod constructors from different sets of information
  • probably but not necessarily fromiter
• copy, deepcopy with the same overloads as the constructor
• empty_like, zeros_like, ones_like with overrides passed to copy
• read/write properties for all representations
• dtype for the nearest Numpy approximation (used by __array__ when that's enabled)
• __len__, shape for the Numpy shape
• type for the awkward Type
• base for resolving views
• _valid for any checks before an access
• _argfields for checking arguments in a higher-order function
• __iter__ for iteration (optimization over __getitem__)
• __getitem__
  • string-slice pass-through to Table
  • empty tuple returns self
  • non-tuples promoted to tuples, with head...
  • integer index returns an item
  • slice returns an array of the same type
  • mask array returns an array of the same type
  • index array returns an array of the same type
  • with tail passed through
• __setitem__ and __delitem__ pass-through to Table
• __array_ufunc__ handles Numpy ufuncs
• any and all
• concat (classmethod) concatenates
• columns, allcolumns pass-through to Table
• pandas makes a DataFrame

The superclass defines (subclasses may override):

• __repr__, __str__ for standard representation
• __array__ which is currently raising an exception so that I catch performance bugs
• __iter__ if no optimization over __getitem__ is needed
• __bool__/__nonzero__ to complain about ambiguous truth value
• tolist converts arrays to lists, Numpy numbers to Python numbers, Table.Rows to dicts
• valid returns True/False based on whether _valid raised an exception
• higher-order functions
  • apply
  • filter
  • maxby
  • minby