Linking and linker relaxation fixes

[DNedic]

This PR fixes the following issues:

1. Accounts for the possibility of the .sdata* sections being generated. These sections are created by compilers in order to optimize access for small data on some architectures. Right now this breaks the espressif/esp example the .sdata output section is being automatically placed at the wrong place, creating an incorrect and enormous binary.
2. Properly provide the __global_pointer$ used in the RISC-V architecture CPUs to initialize the GP register. This register is then used in order to improve access times to frequently used data, as long as it is +-2048 bytes away by fitting the offset in the immideate part of the instruction.
3. Disables relaxation when initializing the GP register on espressif-riscv

References: https://www.sifive.com/blog/all-aboard-part-3-linker-relaxation-in-riscv-toolchain https://five-embeddev.com/quickref/global_pointer.html https://d3s.mff.cuni.cz/files/teaching/nswi200/202324/doc/riscv-abi.pdf

[DNedic]

The toolchain does not appear to generate gp relative loads still, does zig utilize this feature?

[DNedic]

It does appear that relaxation is not enabled in zig: https://github.com/ziglang/zig/issues/3451, I believe this should still be merged now as espressif/esp example (and other projects utilizing espressif-riscv) are broken.

[mattnite]

@DNedic thank you!