Preserve equals signs in query filter values

[harbinresearcher]

Fixes #60.

The CLI query tool rejects filter values containing equals signs, even though query_app() matches the same string. For example, a directive registered with name='https://example.test/?q=python' cannot be selected with decq foo 'name=https://example.test/?q=python': splitting on every = raises ToolError: Cannot parse query filter, no =..

Split at the first equals sign and retain the remainder as the value. Keep whitespace stripping, missing-separator errors and the conversion/matching pipeline. Add a query-tool docstring note, changelog entry, seven parsing cases, and a public CLI regression checking both the matching and nonmatching registrations. All eight new cases fail before the fix.

Validation:

• Both pristine and patched Linux trees pass the project's tox -e py311,py314,coverage,lint,docs,mypy,pyright: full tests, 100% statement coverage, Ruff check/format, strict Sphinx HTML/doctests, and the configured 3.11-3.15 targets of both type checkers.
• Linux CPython3.11.17 and3.14.8: full suite137 ->145 passed; configured statement coverage remains100% (741/741).
• Windows CPython3.12.14: targeted query-tool suite35passed. Complete suite132 ->140passed with the same five pre-existing failures in test_error.py: those assertions expect /test_error.py but Windows reports backslashes. Failure identities were compared; configured statement coverage is100% on both trees. No claim that the Windows full suite passes.
• Windows Ruff check/format, mypy and pyright pass.
• Isolated sdist/wheel build succeeds; the separately installed wheel passes all35 query-tool tests with its actual import path verified.
• Independent automated read-only review found no actionable issues, reran the35 target tests and probed364 filter strings plus generator/duplicate-key inputs. This is not human approval.

PyPy, macOS, and the remaining runtime matrix were not run locally. This is a reproduction using ordinary string attributes, with no downstream production-incident claim.

Disclosure: investigation, implementation, testing, and independent review were performed autonomously with AI under the contributor's authorization. Maintainer review is still needed.