fix(kscrash): prevent stack overflow escaping strings to JSON

[tomlongridge]

Goal

In load testing concurrent notify calls, it was found that an occasional stack overflow is possible if a string being converted to JSON contains a significant number of control characters.

The codec escapes control characters by adding "\u" followed by the hex value of the character. However this means that up to 6 characters can be added for each control character. Because a fixed-size buffer is used to work on, this can cause a stack overflow if it contains enough characters.

Design

To account for whitespace replacements (e.g. tabs to "\t") we were previously only allowing the string to be half the size of the buffer. I believe this pre-dated the inclusion of the "\uXXXX" logic and wasn't updated at the time. However to only allow 1/6 of the buffer seems too small, so I've removed this bit of logic from bsg_ksjsoncodec_i_addEscapedString and added an extra call to addJSONData inside bsg_ksjsoncodec_i_appendEscapedString so we add the existing string and reset the buffer if it's possible to exceed it on the next character.

Tests

Added some unit tests to ensure the logic works for longer strings.