improve implementation of Si and JN (Trac #1193)

[pkienzle]

sasmodels implementation of Si and JN are very badly behaved across much of their range.

Migrated from http://trac.sasview.org/ticket/1193

{
    "status": "new",
    "changetime": "2018-11-15T16:25:24",
    "_ts": "2018-11-15 16:25:24.117589+00:00",
    "description": "sasmodels implementation of Si and JN are very badly behaved across much of their range.",
    "reporter": "pkienzle",
    "cc": "",
    "resolution": "",
    "workpackage": "McSAS Integration Project",
    "time": "2018-10-12T20:17:10",
    "component": "SasView",
    "summary": "improve implementation of Si and JN",
    "priority": "major",
    "keywords": "",
    "milestone": "sasmodels 1.0",
    "owner": "",
    "type": "defect"
}

[pkienzle]

Trac update at 2018/11/15 16:25:24: pkienzle commented:

The cephes implementation of Si is more accurate than that which we are using in sasmodels. Consider using it instead.

[pkienzle]

Use explore/precision.py to see the difference. For example:

$ python explore/precision.py Si

produces:

Similarly for J_{-3}, J_3 and J_2:

$ python explore/precision.py J-3
$ python explore/precision.py J3
$ python explore/precision.py J2