fix: use removeprefix to decode DynamoDB namespace property keys

[anxkhn]

Rationale for this change

DynamoDbCatalog stores each namespace property under a key prefixed with
PROPERTY_KEY_PREFIX ("p.") via _add_property_prefix, and decodes it on read
with _remove_property_prefix:

def _remove_property_prefix(prop: str) -> str:
    return prop.lstrip(PROPERTY_KEY_PREFIX)

str.lstrip treats its argument as a set of characters to strip, not a literal
prefix. With PROPERTY_KEY_PREFIX == "p." it removes every leading p and .
from the decoded key, so any namespace property whose name starts with one of
those characters is corrupted when read back through load_namespace_properties:

stored property	returned key (before)
provider	rovider
path	ath
partition	artition
ppp	`` (empty)
.hidden	hidden

update_namespace_properties reads the same decoded properties, so it operates on
the corrupted keys too. Property names that do not start with p or . (such as
comment, owner, location) round-trip correctly, which is why the existing
tests, whose fixtures avoid those leading characters, never surfaced the bug.

The fix uses str.removeprefix, which strips only the literal "p." prefix. The
single caller (_get_namespace_properties) already guards each key with
startswith(PROPERTY_KEY_PREFIX), so behavior for every valid key is unchanged
and only the previously corrupted keys are affected. str.removeprefix is
available in Python 3.9+ and this project requires >=3.10.

Are these changes tested?

Yes. Added test_load_namespace_properties_with_prefix_char_keys in
tests/catalog/test_dynamodb.py, a create_namespace / load_namespace_properties
round trip (backed by moto's mock_aws) over keys provider, path, ppp,
.hidden, and comment. The test fails on the previous lstrip implementation
(keys come back corrupted) and passes with removeprefix.

tests/catalog/test_dynamodb.py passes (40 tests), and make lint (ruff,
ruff-format, mypy, and the other pre-commit hooks) is clean. The DynamoDB catalog
has no separate integration test path; behavior is covered by the mocked unit test.

Are there any user-facing changes?

Yes: a bug fix. Reading namespace properties from a DynamoDB-backed catalog now
returns property keys that begin with p or . correctly instead of truncating
their leading characters. No public API changes.

[jayceslesar]

Do we even need tests? I don't think it is our job to test that the removeprefix builtin works.

Additionally, this abstraction over a builtin is only used in one place, would it be simpler to just use removeprefix in the only place that _remove_property_prefix is used and just have a comment explaining why we are calling removeprefix?

[anxkhn]

@jayceslesar updated as suggested: removed the one-use helper and builtin-focused tests, then called removeprefix directly with a comment explaining the difference from lstrip. rebased onto current main, and the dynamodb tests and all hooks pass locally. could you take another look?