ci(linux): bring up redis-server and mysql-server for unittests

[rajvarun77]

What

Install redis-server and mysql-server in the clang-unittest and
clang-unittest-asan jobs (Linux), right before run_tests.sh, so the backend
integration tests in the unit-test suite execute against a live server instead
of silently no-op'ing.

Why

test/run_tests.sh runs brpc_redis_unittest, which forks a real
redis-server when the binary is found and otherwise short-circuits each
redis-dependent case with a bare return. gtest counts that return as
PASS, not SKIP. Because CI never installed redis-server, 7 of the 14
RedisTest cases (sanity, keys_with_spaces, incr_and_decr,
by_components, auth, cmd_format, quote_and_escape) reported [ OK ]
while doing nothing — a long-standing false green. The remaining 7 cases are
true in-process unit tests and are unaffected.

Installing the server binary lets these cases run for real. mysql-server is
added alongside redis to bring the same live-backend coverage to the in-flight
MySQL backend work (apache#2093 / apache#3310).

Scope

• Binaries are installed only in the two unittest jobs, not in the shared
  install-essential-dependencies action (which feeds compile-only jobs that
  must stay lean).
• test/run_tests.sh is unchanged — it already builds and runs brpc_*unittest.
• No test source changes in this PR; this is the CI-install half. A follow-up can
  convert the silent return skips to honest GTEST_SKIP() and add a
  require-backend flag so a missing server hard-fails CI instead of skipping.

Test plan

• Validated on a fork run with Actions enabled: redis-server + mysql-server
  install, then run_tests.sh exercises the redis backend tests against the live
  server.

[rajvarun77]

The failing clang-unittest-asan job is a pre-existing test race — not caused by this change

This PR only installs redis-server + mysql-server. That install step passes in every job it was added to (clang-unittest ✅, clang-unittest-asan ✅). The single red job fails inside brpc_redis_unittest, on the 5 tests that talk to a redis the test itself forks on port 6479:

brpc_redis_unittest.cpp:173: Failure
Value of: cntl.Failed()   Actual: true   Expected: false
[E111]Fail to connect ...addr=0.0.0.0:6479...: Connection refused
[  FAILED  ] RedisTest.sanity / keys_with_spaces / incr_and_decr / by_components / auth

Why: RunRedisServer() in test/brpc_redis_unittest.cpp forks redis-server --port 6479 and then waits a fixed usleep(50000) (50 ms) before the first client test connects. On the ASAN runner redis only becomes ready ~82 ms after the fork (same job log: fork ≈ 21:00:06.93 → Ready to accept connections 21:00:07.013), so the first 5 client tests hit Connection refused.

Before this change, which redis-server failed, so those 5 tests were skipped (auto-green). Installing redis simply un-skipped them and exposed the latent 50 ms startup race. The race lives in test/brpc_redis_unittest.cpp, independent of this workflow change.