airy function exhibits inaccuracies for negative inputs

[jacksonvanover]

The airy function appears to exhibit larger and larger inaccuracies for negative inputs with increasing magnitudes when compared to mpmath and gsl's implementations which agree (up to 14 significant digits):

evaluating airy(-1):

• gsl/mpmath give 0.53556088329235207
• jmat gives 0.5355608832923521

evaluating airy(-10):

• gsl/mpmath give 0.04024123848644407
• jmat gives 0.03920869208237318

evaluating airy(-10.039):

• gsl/mpmath give 0.00117947085303331
• jmat gives 0.0001427415208784767

evaluating airy(-10.0392):

• gsl/mpmath give 0.00097859799568954
• jmat gives -0.00005811188429199858

evaluating airy(-10.03923511):

• gsl/mpmath give 0.00094333472274730
• jmat gives -0.00009337169944926118