OPENNLP-1826 : Prevent OOM during Array Allocation

[subbudvk]

Summary

• HeadRules (English) and AncoraSpanishHeadRules (Spanish) parsed the
  tag count field from head rules files with Integer.parseInt() and used
  the result directly as an array size with no bounds check. A crafted model
  file with a count of Integer.MAX_VALUE would trigger an immediate
  OutOfMemoryError during parser model loading.

• Added a bounds check in readHeadRules() in both classes: values outside
  [0, 1000] throw IOException before any allocation.

  Since this is constrained by the size of the POS tagset being used this is already a safe margin and a configurable override may not have benefit.

[rzo1]

• Extract 1_000 into a named constant (e.g. MAX_TAGS) with a short comment justifying the bound. Magic numbers are badly (most of the time).
• The check numTags < 0 only catches parseInt(num) - 2 < 0. If num itself is negative or non-numeric, you get NumberFormatException instead of IOException: wrap the parse or validate num first so callers see a single, predictable failure mode.
• Add an explicit upper-bound test using a value just above 1_000 (e.g. 1_001) in addition to Integer.MAX_VALUE as the boundary is the interesting test case.

[rzo1]

@subbudvk LGTM from my side. Could you create a cherry-pick version targeting 2.x too ?