This restructures parse_string in a way, that you need to keep less
state in your head to understand the code.

This is achieved by:
* only changing the input pointer (current position) at a few places
(not all throughout)
* splitting out the UTF16 handling into a separate function
* renaming the variables so you know what they do without additional
context

This only attaches the parsed linked lists to the items passed to
parse_object and parse_array.

This replaces the bit fiddling involved with calculating a new buffer
size by just multiplying the required length by two, paving the way to a
complete switch to size_t at a later point in time.

Introduces a preprocessor directive that aborts the compilation if the
version numbers in the header don't match with the numbers in the c
file.

This sets the type of an item only if parsing was successful.

This means that in case of failure, the item's type will remain to be
cJSON_Invalid.

This fixes a double free that happens when calling cJSON_Delete on an
item that has been used by parse_string and it failed parsing the
string.

The double free happens, because parse_string frees an alias of
item->valuestring, but doesn't set item->valuestring to NULL.

This might cause slight changes in behavior, but it shouldn't break
anything that is not already broken (for example the original macro
could either return a double, or an integer or whatever depending on if
object is NULL or not.)

Replaces the hand written floating point parser with the C library
implementation.

Simplify error handling using goto fail and improve some names.

Makes the control flow easier to reason about and fixes a few potential
memory leaks.

Makes the control flow easier to reason about and fixes a few potential
memory leaks.

Makes the control flow easier to reason about and fixes a few potential
memory leaks.

Introducing the switch defaults in an earlier commit made UTF-16 \u
sequence handling broken.

Makes type conversions explicit, if they alter a value

If the number is bigger or smaller than the biggest or smallest integer,
the behavior would be undefined.

This commit defines it as saturation behavior.

This is useful to programmatically find out the version of cJSON that
has been used (useful in case of scripting language bindings for
example).

There was a long running discussion here
https://github.com/DaveGamble/cJSON/pull/80 how to provide const
correctness for users of cJSON.

To avoid breaking changes for users of cJSON v1 it was decided to
disable this warning.

pragma was tested with gcc 5.4.0/6.2.1 and clang 3.8/3.9.