Fix JSON parsing bug.

[jseyfried]

Currently, if LLM outputs "C:\\my\\path" (which is correct JSON encoding of C:\my\path), taskgen returns "C:\\\\my\\\\path" (i.e. C:\\my\\path). Also, if LLM outputs "C:\\test\\path", taskgen returns "C:\\\test\\\\path" (not "C:\\\\test\\\\path").

This was caused by parsing strings from JSON using match[1:-1] followed by remove_unicode_escape instead of just json.loads(match).

Script to reproduce:

import taskgen
from taskgen.base import strict_json

LLM_OUTPUT = r"""```json{"###Key###": "\\a\\n\\t\\b"}```"""
print("LLM output:", LLM_OUTPUT)

def chat(*args, **kwargs):
    return LLM_OUTPUT

taskgen.base.chat = chat
print("taskgen output:", strict_json('', '', {"Key": "type: string"}))

Before PR: taskgen output: {'Key': '\\\\a\\\n\\\t\\\\b'}
After PR: taskgen output: {'Key': '\\a\\n\\t\\b'}