Add support for IEEE 754 NaN / infinity

This enables unpacking of 32-bit and 64-bit floating-point NaN,
+infinity and -infinity, to the values `:NaN`, `:inf` and `:"-inf"`, but
does not support packing those atoms back into Messagepack floats.

lib/msgpax/unpacker.ex

@@ -63,6 +63,13 @@ defmodule Msgpax.Unpacker do
       quote(do: <<0xD2, value::32-signed>>),
       quote(do: <<0xD3, value::64-signed>>)
     ] => quote(do: value),
+    # IEEE 754 NaN / infinity
+    [quote(do: <<0xCA, 0::1, 0b11111111::8, 0::23>>)] => :inf,
+    [quote(do: <<0xCA, 1::1, 0b11111111::8, 0::23>>)] => :"-inf",
+    [quote(do: <<0xCA, _sign::1, 0b11111111::8, _fraction::23>>)] => :NaN,
+    [quote(do: <<0xCB, 0::1, 0b11111111111::11, 0::52>>)] => :inf,
+    [quote(do: <<0xCB, 1::1, 0b11111111111::11, 0::52>>)] => :"-inf",
+    [quote(do: <<0xCB, _sign::1, 0b11111111111::11, _fraction::52>>)] => :NaN,
     # Negative fixint
     [quote(do: <<0b111::3, value::5>>)] => quote(do: value - 0b100000)
   }

test/msgpax_test.exs

@@ -108,6 +108,18 @@ defmodule MsgpaxTest do
     assert_format 42.1, <<203>>
   end
 
+  test "ieee 754 32-bit inf / nan" do
+    assert Msgpax.unpack(<<202, 0x7FC00000::32>>) == {:ok, :NaN}
+    assert Msgpax.unpack(<<202, 0x7F800000::32>>) == {:ok, :inf}
+    assert Msgpax.unpack(<<202, 0xFF800000::32>>) == {:ok, :"-inf"}
+  end
+
+  test "ieee 754 64-bit inf / nan" do
+    assert Msgpax.unpack(<<203, 0x7FF8000000000000::64>>) == {:ok, :NaN}
+    assert Msgpax.unpack(<<203, 0x7FF0000000000000::64>>) == {:ok, :inf}
+    assert Msgpax.unpack(<<203, 0xFFF0000000000000::64>>) == {:ok, :"-inf"}
+  end
+
   test "positive fixint" do
     assert_format 0, <<0>>
     assert_format 127, <<127>>