Investigate adding official support for jemalloc on linux

Because java relies on glibc's malloc implementation, allocations off-heap are something of a black box that can make troubleshooting memory growth difficult.

In support cases, we have found and fixed memory leaks by injecting a profiling build of `jemalloc` with `LD_PRELOAD` and using its profiling tools. There is also a general consensus that jemalloc is an improved implementation for long-lived and highly-concurrent applications with a better profile for both improved throughput and reduced overhead, so it would be good to have official support.

Consider adding opt-in support for selecting an available jemalloc on linux hosts.

Acceptance Criteria:
- environment flag like `LS_MALLOC_IMPLEMENTATION=jemalloc`;
- if explicitly set on a linux host and able to resolve the jemalloc libraries, inject them into `LD_PRELOAD` in `lib/logstash.lib.sh` _before_ invoking the jruby subprocess; else log helpfully to stderr and continue without injecting.
- run integration tests (possibly a randomised subset) with jemalloc selected
- provide a straight-forward path for using this feature through docker (notably jemalloc and jemalloc-devel are _not_ available in any enabled-by-default repos for the ubi9-minimal base image)

Additional Considerations:
- We _may_ want to support other allocators in the future, such as Microsoft's mimalloc on Windows or Google's tcmalloc on linux. The opt-in flag should be designed in such a way that it's difficult to end up with a configuration that is ambiguous if and when support for other allocators is added.
- Support on Mac is very difficult and my not be plausible since SIP is enabled by default, preventing libraries from being dynamically injected.

Start by reading lib/logstash.lib.sh and tracing how it invokes the JRuby subprocess and how Linux environment flags are handled. Review the existing integration tests and Docker configuration, including the ubi9-minimal image constraints. Done means an explicit jemalloc selection behaves as specified, failures log to stderr and continue safely, integration coverage runs with jemalloc, and Docker has a straightforward setup path.

Written by the indexing model from the issue text.