tests: add an ASan job running the whole test suite.

[u5surf]

Adds an AddressSanitizer job that runs the whole test suite. Following @zhuizhuhaomeng's note that we could run all the test cases, this covers everything rather than a socket subset.

The key part is building LuaJIT with LUAJIT_USE_SYSMALLOC: without it, LuaJIT's allocator never returns memory to the C allocator, so ASan cannot see use-after-free on Lua objects at all.

What the job does

• Builds nginx with NGX_BUILD_ASAN=1, which ngx-build already supports, and LuaJIT with LUAJIT_USE_SYSMALLOC.
• Preloads the ASan runtime ahead of mockeagain. ASan refuses to start behind another library, and some test files (t/065-tcp-socket-timeout.t, t/131-duplex-req-socket.t) put mockeagain in LD_PRELOAD themselves.
• Allows a longer TEST_NGINX_TIMEOUT than the other jobs, since ASan slows every request down.
• Dumps asan-logs/ on failure. An ASan abort otherwise surfaces only as a connection error from the test harness, which is hard to read.

It finds real defects

Before #2525 landed, this job reported a heap-use-after-free in t/067-req-socket.t TEST 20 — the existing regression test from #2504 — on unmodified master:

==nginx==ERROR: AddressSanitizer: heap-use-after-free
WRITE of size 8
    #0 ngx_http_lua_socket_tcp_finalize  src/ngx_http_lua_socket_tcp.c:4592

That was #2517. With the fix merged, the job is clean.

I hereby granted the copyright of the changes in this pull request
to the authors of this lua-nginx-module project.

[zhuizhuhaomeng]

We clould run all the test cases.

[u5surf]

@zhuizhuhaomeng Done — the job now runs the whole suite.

• 231 files, 25,054 tests, no ASan reports.
• t/146-malloc-trim.t is the one file left out: ASan replaces the allocator, so malloc_trim() returns 0 there.
• The ASan runtime is preloaded ahead of mockeagain, so the files that preload it themselves run too.
• TEST_NGINX_TIMEOUT is higher for this job than for the others.
• The job takes about 25 minutes, against 17–19 for the other jobs.

The remaining red job is t/189-http2-subreq-error-wakeup.t, the flake #2523 addresses; it is unrelated to this change.