feat(berlinmod): streaming benchmark + regenerate the MeosOps facades from master-tracked JMEOS

[estebanzimanyi]

Adds the BerlinMOD streaming throughput benchmark and regenerates the in-repo org.mobilitydb.meos.MeosOps* facades, with the vendored JMEOS.jar + libmeos.so rebuilt from the current MobilityDB master surface via the canonical chain MEOS → MEOS-API catalog → JMEOS (track-master-surface).

• tools/codegen_facades.py emits the 57 MeosOps* forwarder facades from the bundled JMEOS functions.GeneratedFunctions ∩ the streaming-relevance baseline.
• MeosSetSetJoin recovered as source.
• Bundled JMEOS.jar/libmeos.so built from MobilityDB master (4493-function catalog).

Verification

mvn test → 10 tests, 0 failures, 0 errors, 0 skipped.

[JashanReel]

This PR is trying to merge into MobilityDB:consolidate/flink-meos-integration from estebanzimanyi:consolidate/flink-benchmark, but shouldn't it be into MobilityDB:main since this PR (#31) stacks on #30 which was only merged into MobilityDB:main ?

[Ar-mie]

The documentation is not updated to fir the code
In the main() the env.execute(name) is not in a try catch
Maybe for the t.keyBy(x -> 0) in the Q5-continuous it should be specified that all events are in one subtask on purpose
The tests mentioned in the PR description mvn test → 10 tests, 0 failures seem not to include the BerlinMODBenchmark

[estebanzimanyi]

Thanks for the review — all four addressed in 0303853:

1. Docs — the BerlinMODBenchmark Javadoc referenced a --vehicles/--events synthetic mode the code no longer has; realigned it to the actual --csv (required) corpus and removed the dead synthetic label.
2. env.execute not in try-catch — each cell's job is now wrapped in try-catch, so a failing cell is recorded (ERROR row) and the matrix continues instead of aborting.
3. keyBy(x -> 0) in Q5-continuous — added a comment: Q5 (pairwise meet) and Q9 (all-pairs) route the whole stream to a single subtask on purpose (a global, non-parallel view).
4. Benchmark not in the tests — added BerlinMODBenchmarkTest (runs a keyed cell and a keyBy(x -> 0) cell through runCell on a small in-memory corpus), so the benchmark path is now exercised by mvn test (12 tests, 0 failures). Also removed the nine BerlinMODQ*LocalTest drivers that sat under src/main — test-named but never run by surefire, and dependent on an uncommitted sample resource.