[cmath] Only use ISO math functions in avr-libm

[salkinium]

avr-libm does not provide float or long overloads of math functions, so linking against __builtin_sinf resolving to just ::sinf will fail.

I discovered this when compiling the modm math unittests for avr, which uses float types and therefore calls std::sin(float), which the stdlibc++ overloads to use __builtin_sinf instead of just forwarding everything to __builtin_sin.

avr-libm is not very consistent about what overloads it provides though, and I'm not sure if this is the right fix, but it makes the unittests link and pass.

 $ avr-nm path/to/avr-gcc/9.x.y/avr/lib/avr4/libm.a --defined-only # => output cleaned up
acos
asin
atan
atan2
cbrt
ceil
copysign
cos
cosh
exp
fdim
floor
fma
fmax
fmin
fmod
frexp
hypot
inverse
isfinite
isinf
isnan
ldexp
log
log10
lrint
lround
modf
modff
pow
round
signbit
sin
sinh
sqrt
sqrtf
square
tan
tanh
trunc

cc @chris-durand

[salkinium]

(I'll add an example later)

[chris-durand]

The C99 float versions partially exist in the avr libc but are macro defines for the double variants. float and double are both 32 bit floats anyway for avr. I don't know what this __CORRECT_ISO_CPP_MATH_H_PROTO is orignally intended for but if it works it's fine.