[sricharanreddycheruku]

Database.info() leaks its returned bytes object on every successful call.
PyBytes_FromString() already returns a new reference, but the extra
Py_INCREF() leaves one reference behind after the caller discards the result.
Repeated database inspection therefore accumulates memory indefinitely.

Return the existing new reference after freeing the native info string. This
also allows a failed bytes allocation to propagate its NULL result rather than
incrementing a NULL pointer; allocation failure was not injected locally.

Add four bounded allocation regressions beside the existing info test, covering
block, stream, vectored and Chimera databases. Each discards 1,000 info results
and checks allocations traced to the test file. The test preserves an existing
tracemalloc session and allows 4 KiB for bookkeeping. On the original extension,
those calls retain 71–79 KiB on Python 3.12 and 87–95 KiB on free-threaded 3.14.
A separate snapshot reproducer confirms 1,000 retained bytes objects per mode
before the fix and none afterward.

Validation on Linux with source-built wheels and PYTHONMALLOC=debug:

• All four new cases fail before and pass after on Python 3.12.14 and
  free-threaded 3.14.7.
• Free-threaded 3.14 full suite: 34 passed, 4 existing skips; importing the
  extension keeps the GIL disabled.
• Python 3.12 full suite: 35 passed, 2 skips, 1 existing Unicode expression
  lifetime failure addressed separately by Release compile metadata references and propagate ID lookup errors #265. This branch does not include
  that fix.
• C formatting, source Black/Ruff, formatting/lint of the new test addition,
  and git diff --check pass. Existing unrelated formatting/lint issues in
  tests/test_hyperscan.py are unchanged.

Independent branch from current main; no linked issue or API change.