fix: emit _metaptr section writable to avoid text relocations

[Jo5ta]

Problem

A tracepoint compiled into a shared library (a plugin/driver .so) expands _CLLTK_EMIT_META_PTR, which emits absolute pointer pairs into the _clltk_<buffer>_metaptr section. That section was flagged read-only ("a"), so on a PIC/PIE object the dynamic linker must patch a read-only page at load time — a text relocation (DT_TEXTREL). -fPIC cannot avoid it (the relocation is structurally absolute), PIE/hardened targets such as AArch64 reject it at load time, and packaging QA (do_package_qa) fails on it.

Fix

Flag the section writable ("aw"). Its pointers then land in the data segment and are applied as ordinary RELATIVE relocations — matching how the compiler already emits the sibling _clltk_tracebuffer_handler_ptr section. No API or ABI change; consumers just recompile against the updated header.

Verified on x86_64 (gcc) and aarch64 (clang/lld): old flag → DT_TEXTREL; new flag → clean RELATIVE relocations.

Why CI missed it, and the guards added

The failing shape was built in CI, but only ever as a linker warning on a green build — nothing inspected the produced binary. Guards so it cannot return:

• test — a packaging consumer test builds a shared library against the installed headers and asserts the produced .so is free of text relocations and an executable stack (the do_package_qa check). This is the shape the pre-existing consumer test (an executable) never exercised.
• ci — the compiler-compat matrix links a tracepoint-bearing shared object with -z text on every compiler and architecture (incl. arm64/s390x under QEMU), so a text-relocation regression is a hard link error.
• build — the tracing library and example shared libraries link with --fatal-warnings (gated off under sanitizers), so a "creating DT_TEXTREL" linker warning fails the build instead of scrolling past unread.