Offer a new define to avoid int64 conversions

[satur9nine]

The new define PRINTF_AVOID_INT64_TO_FLOAT_CONVERSION when combined with PRINTF_USE_DOUBLE_INTERNALLY ensures that int64 to float conversions are not performed.

[satur9nine]

Firstly thank you for this library! I am using it on an ARM Cortex M7 MCU with hardfloat support, compiling with GCC 13 and using the following defines:

// Completly replace newlib
#define PRINTF_ALIAS_STANDARD_FUNCTION_NAMES_HARD 1

// Exponent support is unneeded for our use case
#define PRINTF_SUPPORT_EXPONENTIAL_SPECIFIERS 0

// Writeback is considered a security risk
#define PRINTF_SUPPORT_WRITEBACK_SPECIFIER 0

// Better performance with float, double is slower
#define PRINTF_USE_DOUBLE_INTERNALLY 0

I noticed in the elf dump the following functions __aeabi_l2f and __addsf3 adding 540 bytes of executable code. The root cause was this code:

scaled_remainder = (abs_number - (floating_point_t) number_.integral) * powers_of_10[precision];
remainder = scaled_remainder - (floating_point_t) number_.fractional;

Which uses int_fast64_t intermediates and converts them back to floating_point_t. This existing code allows for more precision when printing %f but for our needs that amount of precision is not required and not worth the performance and space cost. We are just printing basic logs and prefer high performance and save flash for more important things.

I suspect many people choose this library for the exact same reason so this flag would be appreciated. It is completely opt-in so no existing behavior changes.