Enable Fast Floating-Point reading by default in 3.0

[cowtowncoder]

For Jackson 2.x we have following features disabled by default:

• StreamReadFeature.USE_FAST_DOUBLE_PARSER
• StreamReadFeature.USE_FAST_BIG_NUMBER_PARSER
• StreamWriteFeature.USE_FAST_DOUBLE_WRITER

that is, for FP reading to use default JDK implementation and not one from FastDoubleParser (read) / Schubfach (write). This to minimize any chance of breakage, to require users to opt-in for optimized handling.

Since there hasn't been reports of issues during the time we've had (since 2.14, more than a year), it seems reasonable to assume things work well: so let's enable "fast FP" handling by default for 3.0.
We can still consider similar change for later 2.x, but let's start with 3.0.

EDIT: (2025-04-05)

Due to findings on lack of improvements for writing (equivalent or better code in JDK 21) for newer JDKs, default for StreamWriteFeature.USE_FAST_DOUBLE_WRITER will NOT be changed -- it remains disabled. Users on JDK 17 can then enable it.

[j3graham]

One thought on this change - as of Java 19, an implementation of the Schubfach algorithm is used in the JDK itself, and there have been more improvements since then. Perhaps keeping the default USE_FAST_DOUBLE_WRITER using the JDK should be considered. See https://bugs.openjdk.org/browse/JDK-4511638

[pjfanning]

@j3graham we did not change the write config for exactly the reason you describe.

You can see the v3 code here and its javadoc.

https://github.com/FasterXML/jackson-core/blob/master/src/main/java/tools/jackson/core/StreamWriteFeature.java#L115-L129

[cowtowncoder]

I will update description, good call out @j3graham , @pjfanning.