named tuple iterator, fixes for nested structures and column name handling

[tanmaykm]

• Removed iterators that return Julia structs
• Removed schema creators for Julia structs, and those for Protobuf and Thrift
• Added an iterator RecordCursor that gives out named tuples for records. Did not reuse the old name RecCursor to avoid confusion.
• Fixed handling of nested structures in schema. Nested structures will appear as nested named tuples.
• Used string vectors instead of a delimiter character to represent fully qualified column names. Since we would support arbirtary names now, we can not fix any single delimiter. Column path is important in Parquet because it can contain nested structures, and the same name can appear at different paths. The safest way is to represent the fully qualified path as a vector of path elements. Operations using the iterator will remain oblivious to this though.

Ref: #51, should now pass all the failure cases listed there.

cc: @davidanthoff does this look fine?

Some example interactions:

julia> using Parquet

julia> par = ParFile("booltest/alltypes_plain.snappy.parquet")
Parquet file: booltest/alltypes_plain.snappy.parquet
    version: 1
    nrows: 2
    created by: impala version 1.3.0-INTERNAL (build 8a48ddb1eff84592b3fc06bc6f51ec120e1fffc9)
    cached: 0 column chunks

julia> for rec in RecordCursor(par)
           println(rec)
       end
NamedTuple{(:id, :bool_col, :tinyint_col, :smallint_col, :int_col, :bigint_col, :float_col, :double_col, :date_string_col, :string_col, :timestamp_col),Tuple{Union{Missing, Int32},Union{Missing, Bool},Union{Missing, Int32},Union{Missing, Int32},Union{Missing, Int32},Union{Missing, Int64},Union{Missing, Float32},Union{Missing, Float64},Union{Missing, Array{UInt8,1}},Union{Missing, Array{UInt8,1}},Union{Missing, Int128}}}((6, true, 0, 0, 0, 0, 0.0f0, 0.0, UInt8[0x30, 0x34, 0x2f, 0x30, 0x31, 0x2f, 0x30, 0x39], UInt8[0x30], 45285336301663273581805568))
NamedTuple{(:id, :bool_col, :tinyint_col, :smallint_col, :int_col, :bigint_col, :float_col, :double_col, :date_string_col, :string_col, :timestamp_col),Tuple{Union{Missing, Int32},Union{Missing, Bool},Union{Missing, Int32},Union{Missing, Int32},Union{Missing, Int32},Union{Missing, Int64},Union{Missing, Float32},Union{Missing, Float64},Union{Missing, Array{UInt8,1}},Union{Missing, Array{UInt8,1}},Union{Missing, Int128}}}((7, false, 1, 1, 1, 10, 1.1f0, 10.1, UInt8[0x30, 0x34, 0x2f, 0x30, 0x31, 0x2f, 0x30, 0x39], UInt8[0x31], 45285336301663333581805568))

julia> rc = RecordCursor(par)
Record Cursor on booltest/alltypes_plain.snappy.parquet
    rows: 1:2
    cols: id, bool_col, tinyint_col, smallint_col, int_col, bigint_col, float_col, double_col, date_string_col, string_col, timestamp_col

julia> rec, state = iterate(rc);

julia> rec
NamedTuple{(:id, :bool_col, :tinyint_col, :smallint_col, :int_col, :bigint_col, :float_col, :double_col, :date_string_col, :string_col, :timestamp_col),Tuple{Union{Missing, Int32},Union{Missing, Bool},Union{Missing, Int32},Union{Missing, Int32},Union{Missing, Int32},Union{Missing, Int64},Union{Missing, Float32},Union{Missing, Float64},Union{Missing, Array{UInt8,1}},Union{Missing, Array{UInt8,1}},Union{Missing, Int128}}}((6, true, 0, 0, 0, 0, 0.0f0, 0.0, UInt8[0x30, 0x34, 0x2f, 0x30, 0x31, 0x2f, 0x30, 0x39], UInt8[0x30], 45285336301663273581805568))

julia> rec.id
6

julia> rec.bool_col
true

julia> rec.tinyint_col
0

julia> colnames(par)
11-element Array{Array{String,1},1}:
 ["id"]
 ["bool_col"]
 ["tinyint_col"]
 ["smallint_col"]
 ["int_col"]
 ["bigint_col"]
 ["float_col"]
 ["double_col"]
 ["date_string_col"]
 ["string_col"]
 ["timestamp_col"]

Will update the readme and examples. Will also do some more tests and maybe some refactor and cleanup.

[tanmaykm]

I have also put up a PR for the corresponding changes needed to ParquetFiles.jl: queryverse/ParquetFiles.jl#25