perf: Optimize regular numbers parse logics

[filzrev]

This PR intended to optimize performance for number deserialization by changing Parse to TryParse method. It's discussed at https://github.com/aaubry/YamlDotNet/pull/794

Additionally this PR fix double value deserialization problems that are found on UnitTests.

What's Changes

• Replace existing regular number Parse logics with TryParse method.
• Separate float/double parse logics for .NET Framework.
• Add code path for .NET that parse text as double, then checks float value ranges.
• Add related unit tests (UnquotedStringTypeDeserialization_RegularNumbers)

.NET Behavior differences On .NET Framework. float.Parse((double.MaxValue + 1).ToString()) throw OverflowException. But on .NET Core. It returns PosititiveInfinite(∞).

These behavior differences are described at Single.Parse Method page.

In .NET Core 3.0 and later, values that are too large to represent are rounded to [PositiveInfinity] or NegativeInfinity as required by the IEEE 754 specification. In prior versions, including .NET Framework, parsing a value that was too large to represent resulted in failure.

On current implementation. Values outside the range of the float type are converted to -∞/∞. So I've added logics to check float value ranges.