Feature/shared jdt compilation environment

[RadikalJin]

ASTParser.setEnvironment() triggers full classpath scanning (JAR index loading,
type-environment initialisation) on every invocation. With the previous per-file
parse loop each file paid that cost independently, making it O(N) in the number
of files even when the classpath never changes between files.

This change replaces the per-file createAST() calls inside runOperations() with a
single ASTParser.createASTs() batch call that sets up the JDT LookupEnvironment
exactly once for the entire run. Files are collected and content-prefiltered on
the main thread before the batch parse; the resulting CompilationUnit objects are
then dispatched to the existing parallel ExecutorService for concurrent operation
application — so the classpath-amortisation benefit composes cleanly with the
parallel-processing feature.

Key design decisions:

• AstraUtils.createBatchParser() encapsulates the shared-environment parser setup
  (mirrors createParser() but omits per-file settings: setSource, setUnitName,
  setKind, which are invalid / ignored in batch mode).
• batchParseFiles() in AstraCore calls createASTs() via FileASTRequestor and
  returns an absolute-path → CompilationUnit map; the environment is initialised
  once regardless of how many files are in the batch.
• applyOperationsAndSaveWithPreParsedCU() applies operations on the already-parsed
  CU without any additional setEnvironment() call, then runs import cleanup.
• Content-prefiltered files have no-op futures submitted so they still count toward
  the progress denominator, preserving existing progress-reporting behaviour.
• Files that could not be read before batch parsing surface as RuntimeException
  futures, matching the existing per-file error-surfacing contract.
• Thread safety: createASTs() resolves all bindings eagerly during the sequential
  parse phase; by the time worker threads read those bindings the LookupEnvironment
  is no longer being written, so concurrent reads are safe.