post beta2 fixes

midb.asm:
exception handler for peripheral 68881 writes into location 0
This is now fixed

midb.asm:
fcmp2fp was defined as $5428, it should be $5438, this has no
material affect since we use a software comparison algorithm anyway.

library.asm
If library in use, call to Expunge has no affect now.

Fixed up Expunge/Close to work properly. Also saved a few bytes
of code.

Expunge now closes the mathieeedoubbas library when it really expunges.
This will allow that library to also expunge.

May 27,1988
Expunge did not set expunge on last close bit if call to Expunge failed
because the library was opened by others.

May 27,1988
Expunge now protected by Forbid/Permit. Needed because Close would
jump into it and Expunge dicks with unprotected data structures.

May 27,1988
If a 68020/68881 was installed this library would initialize the rounding
and exception control modes. This is done by the basic library itself
when it is opened. This code was removed from this library and it now
assumes the OpenLibrary("mathieeedoubbas.library") will do the right thing.

June 13,1988
May 27,1988 introduced bugs. Permit can destroy d0, so moved the loading
of d0 to after the Permit call.
Also the Ext functions were improperly returning seglist when they should
have been returning d0=#0

Nov 5,1988
library.asm: Did not cal the alternate dbltrans routine if the flag
was set, but called the dblbas routine.

Nov 5,1988
small changes
removed unused code in ieee68881.asm  'line 175 or there abouts'
removed one line in midb.asm that was not needed (commented it out)
removed some tst.w 123456 instructions in sqrt.asm and softieee.asm

Jan 25, 1989
Changed .fd file for DPSincos
it was (double,pf2)  should be (pf2,double)

There should be a new sqrt routine in here now. It passes rounding however
it may not deal with denormalized numbers properly.

4/19/89
convert.asm
missing clr.w d0, was giving bad results for conversion from single
ieee to dbl ieee.

5/24/89
Fixed sqrt.asm for nonnormalized input, it now passes paranoia again.

5/30/89
Now returns NAN for sqrt(x<0.0)  sqrt(-0.0) now returns +0.0
Did some loop unrol optimizations after experimenting with
the single precision sqrt and seeing the results.
sqrt is now 10% faster then previous release.
