Add ETL standardization pipeline for source-agnostic bibliographic data (Base Level)

[kunalg2909-debug]

Overview

This PR implements a robust ETL (Extract → Transform → Validate → Load) pipeline
that makes Bibliometrix-Python source-agnostic, addressing the fragility described
in the project brief where analytical functions only worked reliably with
Web of Science data.

Architecture

• Extract: Reuses the existing parsers.py (WoS/.txt, PubMed/.txt, Cochrane/.txt)
  and pandas readers already present in the codebase — no duplication.
• Transform: Reuses format_functions.py's existing per-source, per-column
  mapping logic (the "Dispatcher"), which already converts each source's
  proprietary column names into the standard WoS field-tag schema.
• Validate (www/services/validation.py, new): Enforces the Type Contract
  required by the spec — multi-value fields must be list[str], scalar fields
  must be str, TC must be int, and no NaN/None may remain in the final
  output. Includes coerce_to_contract(), a non-destructive auto-fix pass that
  repairs common representation issues (e.g. a stray string where a list was
  expected) before re-validating.
• Load (www/services/get_data.py, new): The single entry-point function
  that was referenced in app.py's "Start" button handler but never defined
  anywhere in the codebase. Orchestrates Extract → Transform → Validate →
  Coerce → SR generation → df.set().
• Calculated Fields (SR): Reuses the existing metaTagExtraction(df, Field="SR")
  function from metatagextraction.py rather than reimplementing SR-generation
  logic, per the project requirements.

Bugs Found and Patched

1. format_functions.py / process_single_file() — deletes AU or AF
   entirely based on the user's author-format selection, instead of leaving an
   empty list, violating the always-present-column contract. Patched
   non-destructively in get_data.py by restoring the dropped column as an
   empty list[str] column after parsing.

2. functions/get_maininformations.py — assumed PY (Publication Year) was
   numeric (data["PY"].max() - data["PY"].min()), but the standardized schema
   stores PY as str per spec. Patched to cast PY to int locally, on a
   copy of the working DataFrame, without mutating the shared reactive state.

3. functions/get_maininformations.py — reassigned data = df.get() after
   calling metaTagExtraction(), discarding all previously computed columns in
   the same function (Min_Year, Max_Year, CAGR, the PY cast). Patched to merge
   only the newly derived AU_CO column into the existing local DataFrame.

4. Various format_xx_column() functions — inconsistent typing across
   sources (TC as string instead of int for WoS; VL not fully unwrapped;
   PubMed leaves OA/SC in non-list form). Handled defensively in the Load
   phase via coerce_to_contract() rather than patched at the source, per the
   Phase 5 validation design — documented as a known upstream inconsistency.

Testing / Execution Evidence

See execution_evidence.ipynb for a full, restart-and-run-all-clean notebook
demonstrating:

• Standalone pipeline execution (Extract → Validate → Coerce → SR) against a
  real Web of Science .txt export and a real PubMed .txt export (10 records).
• Screenshots of the live Shiny dashboard successfully running Main
  Information (WoS), Most Frequent Words (PubMed), and Co-occurrence
  Network (PubMed) — proving the standardized data is fully usable by
  existing analysis functions, not just correctly shaped in isolation.

Files Changed

• www/services/get_data.py (new) — pipeline entry point
• www/services/validation.py (new) — validation + coercion module
• www/services/__init__.py — exposes the two new modules
• functions/get_maininformations.py — 2 patches (see above)
• execution_evidence.ipynb (new) — test evidence notebook
• sample_wos.txt, pubmed-bibliometr-set.txt (new) — sample data used in testing
• screenshots/ (new) — dashboard screenshots referenced in the notebook