Datasets: finish pandas 2 freq alias migration in _tsf_reader

[shaun0927]

Issue #, if available: follow-up to #3232

Description of changes:

#3232 replaced a few of the pandas offset aliases that became deprecated in pandas 2.2 (e.g. H → h, T → min), but _tsf_reader.py still emits the remaining deprecated ones: S, M, Q, Y. On pandas 2.3 they produce FutureWarnings, and they are scheduled to stop working in pandas 3.0.

>>> import pandas as pd; pd.__version__
'2.3.3'
>>> from gluonts.dataset.repository._tsf_reader import BASE_FREQ_TO_PANDAS_OFFSET as M
>>> M["monthly"], M["quarterly"], M["yearly"], M["seconds"]
('M', 'Q', 'Y', 'S')
>>> pd.tseries.frequencies.to_offset("30S")
FutureWarning: 'S' is deprecated and will be removed in a future version, please use 's' instead.
>>> pd.tseries.frequencies.to_offset("1M")
FutureWarning: 'M' is deprecated and will be removed in a future version, please use 'ME' instead.

The mapping is the single source used by frequency_converter, which is called by datasets.get_dataset → _tsf_datasets.py when loading any dataset derived from forecastingdata.org (tourism_monthly, tourism_quarterly, tourism_yearly, cif_2016, m5, etc.).

Changes:

• src/gluonts/dataset/repository/_tsf_reader.py: migrate the four remaining aliases to the pandas-2-compatible spellings (s, ME, QE, YE).
• test/dataset/test_tsf_reader.py: update the expected values in test_frequency_converter accordingly, pinning the new contract.

test/dataset/test_tsf_reader.py ...............                          [100%]
============================== 15 passed in 0.13s ==============================

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

Please tag this pr with at least one of these labels to make our release process faster: BREAKING, new feature, bug fix, other change, dev setup