addr2line: enhance call-stack symbolicator, samples, tests, and CI

[TianlongLiang]

Refactor test-tools/addr2line/addr2line.py to reliably symbolicate WAMR call stacks across the wasi-sdk versions WAMR supports (29+), and add test infrastructure plus two end-to-end debug-tools samples that exercise the workflow.

addr2line.py:

• Detect wasi-sdk clang version at startup and dispatch to a "modern" resolver (clang >= 22, single llvm-symbolizer call) or a "legacy" resolver (clang < 22, with a DW_TAG_subprogram interval-table overlay built from a single llvm-dwarfdump --debug-info pass at startup). The overlay fixes the clang-22 wasm DWARF bug where the symbolizer returns the wrong outermost function name (e.g. recurse reported as free) for addresses overlapping wasi-libc declarations.
• Inline frames are rendered with explicit "(inlined into )" annotations so the call chain reads top-down.
• --no-addr (function-name lookup) prefers non-sysroot candidates so short user-code names (a, b, c) don't collide with wasi-libc helpers in the DWARF.
• -v/--verbose gates the dispatch-decision log; default silent.
• Require llvm-symbolizer; drop the legacy llvm-addr2line fallback (every supported wasi-sdk ships symbolizer).

samples/debug-tools (refresh):

• Use always_inline trap_helper to demonstrate inline expansion.
• symbolicate.sh is self-locating and prints banner headers per mode (interp / --no-addr / aot) so the three output blocks are distinguishable.
• verify.sh asserts the precise frame names and inline annotation using anchored regex.
• Fix make -j race in wasm-apps/CMakeLists.txt (explicit add_dependencies for the AOT custom target).

samples/debug-tools-optimized (new):

• Production-realistic workflow: clang -Oz -g -flto then wasm-opt -Oz -g to produce a stripped .prod.wasm plus a DWARF companion (.debug.wasm) with byte-identical code.
• AOT path via wamrc --enable-dump-call-stack --bounds-checks=1.
• WAMR_DISABLE_HW_BOUND_CHECK=1 so OOB traps go through the interpreter exception path (capturing frame_ip) instead of the SIGSEGV signal handler (which doesn't update frame->ip).
• Two apps: oob (cross-TU inlined memory trap) and stackoverflow (non-tail recursion that exhausts the operand stack at ~20 frames with --stack-size=4096).
• verify.sh covers all (app, mode, interp) combinations.

test-tools/addr2line/tests (new):

• Pytest suite covering 13 scenarios: simple resolution, basic and deep inline chains, cross-TU LTO inlining, mid-function trap, multi-frame call stack, C++ demangling, AOT mode, fast-interp fallback, --no-addr, offset=0 fallback, empty input, version dispatch, and modern-vs-legacy equivalence under --multi-sdk.
• conftest.py exposes wasi_sdk/wabt/binaryen/build_wasm fixtures; filters SDK candidates by llvm-symbolizer presence; prints invocation IO under pytest -v -s.
• 8 purpose-built C/C++ test apps under apps/, 3 fixture call stacks.

CI:

• Per-PR job: addr2line_tests removed; debug-tools-optimized folded into build_samples_others (no separate runner / runtime install). Matrix-style classic+fast interp coverage via in-step loop.
• Nightly: addr2line_tests_multi_sdk parametrizes pytest over wasi-sdk 29 and 33 for the modern/legacy equivalence test.
• macOS workflow: existing debug-tools step now calls verify.sh.
• install-wasi-sdk-wabt action: bump default wasi-sdk 25 -> 29 (pre-29 bundles lack llvm-symbolizer). New wasi_sdk_version input lets callers pin a version when needed.

Documentation:

• New test-tools/addr2line/README.md explains the script's architecture (modern/legacy split, the LLVM symbolizer wasm bug, --mode={interp,aot,fast-interp} offset spaces, the four tools involved). Sample READMEs link there to avoid duplication.
• tests/README.md describes every test case with the regression class it guards against.

[TianlongLiang]

This is all in one PR; I will break it into a series of PRs so it is easier to review