mps-cli-py: complete binary (.mpb) persistency implementation

[Prithvi686]

Added full binary (.mpb) model persistency support

MPS stores models in three formats: XML (.mps), file-per-root (.model directories), and binary ('.mpb'). The first two were already supported. This PR adds complete support for the binary format.

Newly Added:
SModelBuilderBinaryPersistency.py: Top-level .mpb parser that parses header, registry, model properties, node tree.

binary/registry.py: Registry section parser that populates 'index_2_concept', 'index_2_property', 'index_2_reference_role', 'index_2_child_role_in_parent', 'concept_id_2_concept'

binary/nodes.py: Node tree parser that containing methods 'read_children', 'read_node', '_read_reference', '_read_node_id'

binary/node_id_utils.py: NodeIdEncodingUtils class that encodes/decodes MPS node IDs between raw long and Base64-variant strings

Modified:
SSolutionBuilder.py: 'build_all()' now processes '.mpb' files in parallel via 'ProcessPoolExecutor' and also added 'USE_CACHE', 'CACHE_LOAD_FN', 'CACHE_SAVE_FN' hooks.

SSolutionsRepositoryBuilder.py: Three performance optimisations:

• before extracting a jar, the zip central directory is peeked to check for any '.msd' file.
• The remaining JARs (those with .msd) are extracted and parsed concurrently via a 'ThreadPoolExecutor' where JAR extraction is I/O-bound, so threads are the appropriate primitive here and no per-process spawn overhead.
• Within each solution, .mpb files are parsed by worker processes thru 'ProcessPoolExecutor' rather than threads and parsing is cpu-bound (binary decoding, string table lookups), and the 'MPB_PARALLEL_THRESHOLD = 4' guard skips the pool for small batches and parses them serially and also solutions in a single JAR share one pool creation, so the cost is paid at most once per jar file.
• After the first parse, each 'SModel' is saved to '~/.mps_cli_cache/' directory keyed by 'md5(path, time, size)' and on subsequent runs files whose path and modification time and size are unchanged are loaded from this directory and the cache is invalidated automatically when a file changes so no manual invalidation is needed but the tests always set 'USE_CACHE = false' to ensure fresh parses

demo.py: parses a plugins directory or test project, prints a structured summary of all solutions/models/nodes, runs a verification pass, and writes output to a timestamped log file

What is parsed and stored in SModel:
Every .mpb file becomes one 'SModel' containing:

• 'uuid' which is Java-style uuid string 'r:xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx' matching 'UUID.toString()' in Java
• 'name' which is a fully qualified model name (ex: 'jetbrains.mps.vcs.diff')
• 'root_nodes' which is a list of 'SNode' objects, each carrying:
  • 'uuid': Base64-variant encoded node ID
  • 'concept': 'SConcept' with the fully qualified concept name resolved from the registry
  • 'role_in_parent': the containment link name this node fills in its parent
  • 'parent': back-reference to the containing 'SNode' (means none for root nodes)
  • 'properties': 'dict[str, str]' of all string-valued property name and value pairs
  • 'references' — 'dict[str, SNodeRef]' mapping reference role name to 'SNodeRef', which carries 'model_uuid', 'node_uuid', and 'resolve_info'
  • 'children' — ordered list of child 'SNode' objects (recursive)

9 new test files (~75 new test methods) have been added to verify all the parsing scenarios.