fix(oxml): keep generated ids within signed int32 range

[seonghoony]

Closes #34.

Root cause

_paragraph_id, _object_id, and _memo_id in src/hwpx/oxml/document.py masked uuid4().int with 0xFFFFFFFF. Since uuid4().int is uniformly distributed over 122 bits of entropy, the masked output is uniform over the full unsigned 32-bit range; about 50% of generated ids have bit 31 set (i.e. >= 2^31).

Downstream consumers that parse the id attribute as a signed 32-bit integer interpret those values as negative, which has caused interop issues.

Fix

Mask one bit lower (& 0x7FFFFFFF) so the result fits in [0, 2^31) and survives any consumer that uses signed int32. Three call sites updated:

 def _paragraph_id() -> str:
     """Generate an identifier for a new paragraph element."""
-    return str(uuid4().int & 0xFFFFFFFF)
+    return str(uuid4().int & 0x7FFFFFFF)


 def _object_id() -> str:
     """Generate an identifier suitable for table and shape objects."""
-    return str(uuid4().int & 0xFFFFFFFF)
+    return str(uuid4().int & 0x7FFFFFFF)


 def _memo_id() -> str:
     """Generate a lightweight identifier for memo elements."""
-    return str(uuid4().int & 0xFFFFFFFF)
+    return str(uuid4().int & 0x7FFFFFFF)

Collision risk is essentially unchanged — birthday probability for N=10_000 ids drops from ~2.3e-5 (32-bit) to ~4.7e-5 (31-bit), which is still negligible for any realistic document size.

Regression test

tests/test_id_generator_range.py (new):

• test_id_generators_stay_within_signed_int32 — samples 200 values from each of the three generators and asserts 0 <= v < 2**31.
• test_id_generators_use_full_31_bit_range — samples 4,000 values from each generator and asserts at least one is >= 2^30, guarding against accidental over-restriction (e.g. a future change that masks too many bits).

Without the fix, test_id_generators_stay_within_signed_int32 fails immediately:

AssertionError: _paragraph_id produced 3521445198 (0xd1e4fd4e); must be in [0, 2^31)

With the fix:

$ pytest tests/test_id_generator_range.py
tests/test_id_generator_range.py ..                                      [100%]
============================== 2 passed in 0.06s ===============================

Wider regression check

$ pytest tests/test_id_generator_range.py tests/test_oxml_parsing.py tests/test_document_save_api.py
14 passed, 2 skipped in 0.12s

(2 skips are pre-existing, due to absent sample HWPX fixtures in this checkout.)

Notes

• This PR is intentionally minimal: only the three random ID generators are touched.
• The _allocate_*_id helpers (_allocate_char_property_id, _allocate_border_fill_id, _allocate_bin_item_id) propagate max(numeric_ids) + 1. They will start producing in-range values once the input document also has in-range ids — that is a follow-up topic and not part of this PR.
• src/hwpx/data/Skeleton.hwpx ships with <hp:p id="3121190098"> (>= 2^31). That is filed separately as Skeleton.hwpx ships with <hp:p id="3121190098"> (out of signed int32 range) #35 and patched separately so the two fixes can land independently.