AP_Math: Log source line of constrain_float nan's

[WickedShell]

This makes tracking down the source of constrain nan internal errors much easier. This change causes us to log the line number where we called constrain_float rather then the line number inside of the constrain_float function, which was never very helpful, as far to many places call that.

This change does try and minimize the flash cost for small boards that are marginal, and only add's the line data on boards with more then 1MB of flash. This is a bit debatable to me, as the utility of this is so high that I think we should probably always be logging this, as if you hit this at the moment on a 1MB board it's very very difficult to diagnose what happened, whereas with this we at least have a good chance to track it down.

This whole PR has been used in anger to help actually track down the source of some constrain_nan errors in real code.

CubeOrange stats:

Build	Target	Text	Data	BSS	Total
This PR	bin/arducopter	1485608	1520	129760	1616888
Master	bin/arducopter	1484016	1520	129760	1615296
Delta		1592	0	0	1592

[WickedShell]

omnibusf4pro stats:

Build	Target	Text	Data	BSS	Total
Enabled	bin/arducopter	941968	2620	128536	1073124
Disabled	bin/arducopter	940844	2620	128540	1072004
Delta		1124	0	-4	1120