[bot] Merge master/8c525d1a into rel/dev - #1953

gooddata-pandas capped pandas below 3, which kept users who also use
pandas directly on pandas 2, and pandas 2 publishes no CPython 3.15
wheels (2.3.3 built from source segfaults inside pd.to_datetime on
3.15). The pin now allows pandas >=2.0.0,<4.0.0, and Python 3.15
requires pandas 3.

Existing environments keep their pandas 2; new ones resolve pandas 3,
whose behavior changes reach users through the objects gooddata-pandas
returns: string columns and index levels use the default "str" dtype,
missing values in string columns stay missing, and copy-on-write is
always on. The release notes must call this out.

pandas 3 infers datetime resolution (dates become datetime64[s]) where
pandas 2 always returned datetime64[ns]. Date and datetime attribute
conversion in the JSON and Arrow paths goes through to_datetime_ns(),
which yields datetime64[ns] on both majors; on pandas 2 the as_unit
call is a no-op.

On pandas 3, an index built from Arrow field metadata (transposed
results, or no row labels) gets the default "str" dtype instead of
object; it is never Arrow-mapped. The ARROW_STRINGS test now checks
index dtypes only when the index comes from the inline row-label
columns, so a row-label level that loses its Arrow-backed dtype still
fails it.

The lock resolves pandas 3 on Python 3.11+, so CI tests pandas 3
there; pandas 2 is covered only by the Python 3.10 environment.

jira: trivial
risk: low

Adds the 3.15 classifier to every published package and py315 to the
tox envlists, the CI matrix and ci_tests.mk. Python 3.10 stays
supported.

The py315 CI image is python:3.15.2-slim-trixie rather than bookworm:
orjson publishes cp315 Linux wheels only for glibc 2.39+, and bookworm
ships glibc 2.36, so uv falls back to the sdist, which needs a Rust
toolchain the slim image lacks. Trixie ships glibc 2.41.

Sonar's python.version also gains 3.14, which it was missing.

jira: trivial
risk: low

Python 3.10 reached end of life in October 2026. requires-python moves
to >=3.11 in every package; the 3.10 classifier, tox env, CI matrix
entry and Docker target go away, and the docs list py311-py315. As
with the Python 3.8 (1.25.0) and 3.9 (1.54.0) drops, this ships in a
minor release: pip and uv on 3.10 keep resolving to the last release
that supports it. The release notes must call the drop out.

ruff targets py311, so the safe UP017 fix (datetime.UTC instead of
timezone.utc) and the stdlib import ordering it implies are applied.
UP042 ((str, Enum) -> StrEnum) stays ignored because it changes the
str() and format() output of public enums, which is breaking.

The lock no longer carries Python 3.10-only resolutions (pandas 2.3.3,
pyarrow 25, jiter 0.18, numpy 2.2). With pandas 2 gone from the lock,
CI no longer tests gooddata-pandas against pandas 2.

jira: trivial
risk: low

feat(repo): support Python 3.15, allow pandas 3, drop Python 3.10