fix: avoid amplifying the rounding of x^2 for negative x in math/base/special/erfcx

[Abhist17]

Description

What is the purpose of this pull request?

This pull request:

• evaluates exp(x^2) in the negative branch of math/base/special/erfcx as exp(xh^2) * exp((x-xh)*(x+xh)) (JS and C), where xh is x truncated to 16 fractional bits, so xh*xh is exact and the second factor is close to one.

For x < 0, erfcx(x) = 2 exp(x^2) - erfcx(-x). Forming x*x rounds it, and exp turns that absolute error of about x^2 ε into a relative error of the same size. That is several hundred ULPs once x^2 is in the hundreds:

var erfcx = require( '@stdlib/math/base/special/erfcx' );

erfcx( -25.150567895691 );
// develop: 1.0338731430900656e+275
// this PR: 1.033873143090124e+275 (256-bit BigFloat: 1.033873143090124e+275)

On the new fixture (2001 points in [-26.6, -1]), develop is off by up to 493 ULPs and this PR by up to 2.

• moves medium_negative.json ([-10, -1], same grid) from the SciPy runner to a Julia runner that evaluates erfcx in 256-bit BigFloat. SciPy's erfcx uses the same Faddeeva-package construction, so its values carried the same error, up to 59 ULPs at x = -9.19284876905041 (SciPy 1.005985069797548e37, BigFloat 1.0059850697975548e37). That test is now checked at 2 ULPs instead of 1.
• adds negative.json, down to -26.6, just above where the function overflows (-26.7).

Related Issues

Does this pull request have any related issues?

None.

Questions

Any questions for reviewers of this pull request?

No.

Other

Any other information relevant to this pull request? This may include screenshots, references, and/or implementation notes.

test.js and test.native.js pass.

Checklist

Please ensure the following tasks are completed before submitting this pull request.

• Read, understood, and followed the contributing guidelines.

AI Assistance

When authoring the changes proposed in this PR, did you use any kind of AI assistance?

• Yes
• No

If you answered "yes" above, how did you use AI assistance?

• Code generation (e.g., when writing an implementation or fixing a bug)
• Test/benchmark generation
• Documentation (including examples)
• Research and understanding

[stdlib-bot]

Coverage Report

Package	Statements	Branches	Functions	Lines
math/base/special/erfcx	$\\color{red}1373/1448$ $\\color{green}+94.82\\%$	$\\color{green}95/95$ $\\color{green}+100.00\\%$	$\\color{red}79/104$ $\\color{green}+75.96\\%$	$\\color{red}1373/1448$ $\\color{green}+94.82\\%$

The above coverage report was generated for the changes in this PR.