Fix get_data() dropping files when a DataTableID has multiple URLs

[khider]

This PR fixes #58

Problem

NOAADataset.get_data() assumed a 1:1 mapping between DataTableID and file URL — it only ever fetched paleo_data.file_url, a shortcut to the first file NOAA attached to that table, silently ignoring the rest. In practice, one DataTableID can carry several files of different formats (a NOAA-templated .txt, the original contributed .txt, a .csv, a directory-listing page, or proprietary tree-ring formats), and these aren't always the same table in another format — e.g. a tree-ring DataTableID can bundle a separate "Raw Measurements" table and a "Chronology" table under one ID. The old code returned at most one of them, and would outright crash if the first-listed file happened to be a non-.txt link (e.g. a folder).

Separately, DataTableIDs were assumed unique; if the same ID ever appeared under two different Study/Site pairs, the internal index silently kept only the last one.

Fix

1. pyleotups/core/NOAADataset.py

get_data(dataTableIDs=...) now attempts every file under a DataTableID, ranked easiest-first (NOAA structured .txt → Original Contributed .txt → .csv → web page/folder link via new _rank_file_priority()), and keeps a DataFrame for every file that parses instead of stopping at the first success. A failing file is logged and skipped rather than aborting the rest; it only raises if none of a table's files could be parsed.

Duplicate DataTableIDs (same ID across different Study/Site pairs) are now surfaced via _check_duplicate_datatable_ids(), called from both get_tables() and get_data(). A new data_table_index_all multi-map lets get_data() open files from every colliding Study/Site rather than only the one left in the old single-entry index.

_process_file() gains support for .csv (via pandas.read_csv) and for web pages / directory-listing links (no extension or .html): it fetches the page, extracts links with the stdlib html.parser (no new dependency), ranks them by the same priority, and attempts each — addressing the optional "detect folder, extract URLs" ask.

Proprietary tree-ring formats (.crn, .rwl, .fhx, .lpd) remain explicitly unsupported and are still rejected, just now as one skipped attempt among several rather than a hard stop.

2. pyleotups/tests/test_NOAADataset.py

New tests covering: priority ranking, falling back to the next file when the easiest one fails, raising when all files fail, keeping every DataFrame when multiple files succeed, duplicate-ID warnings in get_tables()/get_data(), opening files from every colliding Study/Site, .csv parsing, and folder-page link extraction/parsing.

Updated one existing assertion (test_get_data_t05_unsupported_file_type_raises) to match the more general "unsupported file type" message now that .csv and web pages are supported types.

Verification

Full suite: 121 passed (117 pre-existing + new tests), no regressions.

Checked against the live NOAA API, not just mocks:
NOAAStudyId=6400: 3 DataTableIDs each with a folder-link + an alternate .txt. Old code crashed on the very first ID (folder link has no .txt extension); fixed code returns all 3 successfully. This study was also added to the tutorials as an example.

xmlId=6400 (tree-ring study, DataTableID 6773, 5 files): now correctly returns 2 DataFrames (Chronology + Raw Measurements, the two NOAA-template .txt files) while clearly warning and skipping the 3 genuinely unparseable files (a prose correlation-stats report, .crn, .rwl).

Use of AI

The fix was generated by Claude (including new test suites). Human reviewed the verification against NOAA API manually.

[khider]

Problem with failure of tests comes from the new Pangaea update and will require new fixes. Merging things for now.