Fix inclusive integer max regressions

[kendonB]

This follow-up fixes the regressions introduced by making Integer-family max bounds inclusive in #395.
Summary:

• keep Integer, IntegerRef, and ShortIntegerRef using inclusive public max bounds
• preserve the caller-facing max value on Integer while still passing an exclusive upper bound to the internal builders
• tighten Number series chunks back to 0..99 / 1..99 so the series form does not start accepting 100 after the inclusive change
• tighten English calendar/time bounds that relied on the old exclusive behavior:
  • years: 2000..2099
  • absolute dates: 1..31
  • relative dates: 1..99 days
  • twelve-hour time: 1..12, minutes 0..59
  • military time: 00..23, minutes 0..59
• add regression coverage for:
  • Integer(min=1, max=2) including the upper bound
  • Number rejecting three-digit series chunks
  • English calendar/time upper bounds
• document the inclusive-range behavior in the language docs and changelog
  Validation:
• text suite passes with the added calendar coverage

assisted by codex gpt5.4 xhigh

[kendonB]

Added a follow-up commit to fix a Repetition.value() regression in the inclusive-max changes.

Problem:
When max == min + 1 (for example Repetition(child) or Repetition(child, min=1, max=2)), recognition succeeds for the maximum count, but get_repetitions() only walked the optional tail when self._max - self._min > 1. That drops the final repetition from the returned value.

Fix:

• change the tail traversal condition in dragonfly/grammar/elements_basic.py from > 1 to > 0
• add a focused text-engine regression test covering the two-item case

I pushed this as commit 9e2b5d9 on review-pr-395.
comment by 5.4 xhigh codex

Checked this one and it looks to be consistent with the idea of the original PR