improved BigDecimal parsing performance

[wenshao]

• Use bit-shift operations for faster digit parsing
• Simplify overflow detection to single check after digit loop
• Handle exponent with combined scale to avoid string-based parsing

What this PR does / why we need it?

Summary of your change

Please indicate you've done the following:

• Made sure tests are passing and test coverage is added if needed.
• Made sure commit message follow the rule of Conventional Commits specification.
• Considered the docs impact and opened a new docs issue or PR with docs changes if needed.

[wenshao]

The same overflow check is duplicated in core/src/main/java/com/alibaba/fastjson2/JSONReaderUTF16.java and should be fixed there as well.

— gpt-5.5 via Qwen Code /review

[wenshao]

[Critical] This post-loop overflow check can accept corrupted 19-digit values. For exactly 19 digits, values greater than Long.MAX_VALUE may overflow during longValue = longValue * 10 + digit and still finish with a non-negative wrapped longValue, so they take the BigDecimal.valueOf(longValue, combinedScale) fast path instead of falling back to TypeUtils.parseBigDecimal(...). That can silently return the wrong BigDecimal for valid JSON numbers outside signed long range.

A safer fix is to restore per-digit overflow detection, or add an exact 19-digit range check before using the accumulated longValue; otherwise force the existing string-based fallback.

— gpt-5.5 via Qwen Code /review