Fixes alignment in ELF shared object templates

[SilentSobs]

The _start label in the ARM LE ELF shared object template was landing
at offset 0xF6, which is half-word aligned but not word aligned.

ARM A32 requires 4-byte (word) alignment the least significant two
bits of the entry point address must always be zero. 0xF6 violates
this, causing the shared object to fail to load and execute on 32-bit
ARM Linux targets when using --format elf-so.

Fix: added 2 null padding bytes between strtab and _start in
elf_dll_armle_template.s, pushing the entry point from 0xF6 to
0xF8 (word aligned). The compiled template_armle_linux_dll.bin has
been regenerated with nasm.

Fixes #19668

Verification:

• Generate an elf-so payload for armle:
  msfvenom --arch armle --platform linux --payload linux/armle/exec --format elf-so -o test.so CMD="id"
• Verify entry point is word-aligned:
  readelf -h test.so → Entry point address: 0xf8 ✅
• Confirmed by original reporter (sfewer-r7) that this fix works on a real 32-bit ARM Linux target (issue ARM32 LE Shared Object template alignment issue #19668)
• No regression on x86 elf-so (entry: 0xf6, correct) and x64 elf-so (entry: 0x192, correct)

[msutovsky-r7]

Also noticed that the same bug might be introduced in riscv32le, so that should be addressed as well.

[msutovsky-r7]

Release Notes

Fixes ELF shared object (elf-so) payload generation failing on 32-bit ARM Linux and RISC-V 32-bit LE targets. The _start entry point in the ARM LE template was landing at a non-word-aligned offset, which violates the architecture's 4-byte alignment requirement and caused the shared object to fail to load. The templates now use proper NASM align directives to ensure correct entry point alignment, and a similar fix is applied to the RISC-V 32-bit LE template.