Unquoted strings starting with non-string token do not strictly preserve its original content

[andjsrk]

Description

Currently, the parser parses numbers immediately in tokenization step then stringifies them again when it turns out to be part of an unquoted string. This should probably be changed to store a string and parse only when it really is a number, in parsing step.

Reproduction Steps

1. Call the decoder with:
   Case 1:

   a: 1.0 b
   

   Case 2:

   a: 1e1 b
   

2. Observe the output.

Expected Behavior

Case 1: Parsed to { "a": "1.0 b" } (JSON for convenience)
Case 2: Parsed to { "a": "1e1 b" } (JSON for convenience)

Actual Behavior

Case 1: Parsed to { "a": "1 b" } (JSON for convenience)
Case 2: Parsed to { "a": "10 b" } (JSON for convenience)

Environment

• Package version: 0.4.4
• Runtime: Rust 1.96.0 (nightly)
• OS: Windows 11

Additional Context

No response

[shreyasbhat0]

Fixed in #62. The scanner now stores the raw number string (e.g. "1.0", "1e1") as last_token_text instead of using n.to_string() which would lose formatting. 1.0 b now correctly parses as "1.0 b" instead of "1 b".