[NO-TICKET] Fix Ruby 2.3 system tests by using Debian 11 bullseye as a base

[ivoanjo]

What does this PR do?

This PR changes our Ruby 2.3 docker image to use Debian 11 "bullseye" as a base instead of Debian 12 "buster".

In #3534 we updated this image but this broke system-tests
(https://github.com/DataDog/dd-trace-rb/actions/runs/8346827939/job/22845592152) because they use an old nokogiri version that fails to build on modern systems (sparklemotion/nokogiri#2105):

$ gem install nokogiri -v '= 1.10.10'

xml_document.c:495:14: error: conflicting types for 'canonicalize'; have 'VALUE(int,  VALUE *, VALUE)' {aka 'long unsigned int(int,  long unsigned int *, long unsigned int)'}
  495 | static VALUE canonicalize(int argc, VALUE* argv, VALUE self)
      |              ^~~~~~~~~~~~
In file included from /usr/include/features.h:489,
                 from /usr/include/x86_64-linux-gnu/bits/libc-header-start.h:33,
                 from /usr/include/stdlib.h:26,
                 from ./nokogiri.h:4,
                 from ./xml_document.h:4,
                 from xml_document.c:1:
/usr/include/x86_64-linux-gnu/bits/mathcalls.h:370:1: note: previous declaration of 'canonicalize' with type 'int(double *, const double *)'
  370 | __MATHDECL_1 (int, canonicalize,, (_Mdouble_ *__cx, const _Mdouble_ *__x));
      | ^~~~~~~~~~~~
make: *** [Makefile:240: xml_document.o] Error 1

make failed, exit code 2

Gem files will remain installed in /usr/local/bundle/gems/nokogiri-1.10.10 for inspection.
Results logged to /usr/local/bundle/extensions/x86_64-linux/2.3.0/nokogiri-1.10.10/gem_make.out

Rather than fighting nokogiri or the system-tests app, let's instead gently step back from Debian 12 to 11, which should still buy us plenty of time until we deprecate Ruby 2.3.

Motivation:

Fix CI/system-tests for Ruby 2.3.

Additional Notes:

N/A

How to test the change?

I've locally built the image with

$ docker build . -f Dockerfile-2.3.8 -t some-testing-name

and confirmed that nokogiri successfully installs.

For Datadog employees:

• If this PR touches code that signs or publishes builds or packages, or handles
  credentials of any kind, I've requested a review from @DataDog/security-design-and-guidance.
• This PR doesn't touch any of that.

[ivoanjo]

Notes on this PR:

1. System-tests will still fail since it won't pick up the new image until we publish it (https://github.com/DataDog/dd-trace-rb/actions/runs/8356497866/job/22873701994?pr=3541)

2. CircleCI tests for 2.1/2.2/2.3 are failing with many different reasons (missing ffi, missing database driver). I believe the fix here is to bust the CircleCI caches -- they are probably caching gems from a different debian base image and they need to be rebuilt. We need to bust the cache on a later PR. I say a later PR because this is a) unrelated to this change in particular; b) we're going to need to bust the cache again after publishing this new image.

The key bit to look out is for https://github.com/DataDog/dd-trace-rb/actions/runs/8356497863/job/22873697379?pr=3541 -- the new image build do succeed.