Remove "--hash-style=both"

[tianon]

It turns out that --hash-style=gnu is considered better than either of --hash-style=both or --hash-style=sysv, assuming your environment/platform supports it.

On amd64 with both Debian's and Alpine's linkers, --hash-style=gnu is the default.

This is especially relevant on MIPS (mips64le), where ld: .gnu.hash is incompatible with the MIPS ABI (so the linker sanely defaults to --hash-style=sysv there, as it should).

Refs #272

[tianon]

Forgot to include the pudding. 😅

Current: (--hash-style=both)

$ docker run --rm php:7.4-alpine sh -c 'apk add --no-cache binutils && readelf --headers /usr/local/bin/php' | grep '[.]hash'
  [ 2] .hash             HASH             0000000000000290  00000290
  [ 3] .gnu.hash         GNU_HASH         00000000000046e0  000046e0
   02     .interp .hash .gnu.hash .dynsym .dynstr .gnu.version .gnu.version_r .rela.dyn .rela.plt 

This PR on amd64 (both Debian and Alpine builds):

$ docker run --rm 60ccb5f28a3c readelf --headers /usr/local/bin/php | grep '[.]hash'
  [ 4] .gnu.hash         GNU_HASH         0000000000000308  00000308
   02     .interp .note.ABI-tag .note.gnu.build-id .gnu.hash .dynsym .dynstr .gnu.version .gnu.version_r .rela.dyn .rela.plt 

$ docker run --rm c22cf30dded9 sh -c 'apk add --no-cache binutils && readelf --headers /usr/local/bin/php' | grep '[.]hash'
  [ 2] .gnu.hash         GNU_HASH         0000000000000290  00000290
   02     .interp .gnu.hash .dynsym .dynstr .gnu.version .gnu.version_r .rela.dyn .rela.plt 

This PR on mips64le:

$ docker run --rm c138082f66eb readelf --headers /usr/local/bin/php | grep '[.]hash'
  [ 5] .hash             HASH             0000000000004cb8  00004cb8
   03     .MIPS.abiflags .MIPS.options .note.gnu.build-id .dynamic .hash .dynsym .dynstr .gnu.version .gnu.version_r .rel.dyn .init .text .MIPS.stubs .fini .rodata .interp .eh_frame .note.ABI-tag 

[tianon]

FWIW, https://flameeyes.blog/2010/09/09/are-we-done-with-ldflags/ has a pretty good discussion of the history behind the two options -- mostly, the reason to use both is so that your final binaries can be used on older userspace implementations than the one you've compiled in, which is definitely not a goal of this image (it's expected that you're compiling things to use within the same variant of the image).