Add ISO 4217 currency column to XBRL facts DataFrame (#850)

[gaoflow]

Summary

Resolves #850. xbrl().facts.to_dataframe() exposed the raw XBRL unit_ref id for every fact. For non-USD filers that id is an opaque token such as UNIT_STANDARD_HKD_MNUSOXGRF0O9R60JINVDUQ instead of a usable currency, which made currency-based filtering and display unreliable for foreign companies (reported by @warzoo).

What changed

The facts DataFrame now carries a currency column with each fact's ISO 4217 code (e.g. USD, HKD), resolved from the unit's iso4217: measure that the parser already records in xbrl.units. The opaque unit_ref is preserved unchanged — currency is purely additive:

df = filing.xbrl().facts.to_dataframe()
df["unit_ref"]   # "UNIT_STANDARD_HKD_..."  (unchanged)
df["currency"]   # "HKD"                    (new)

• Per-share monetary units (e.g. iso4217:USD / xbrli:shares) report their numerator currency (USD).
• Non-monetary units (shares, pure, company-specific) resolve to None rather than a misleading value.

Verification

tests/issues/regression/test_issue_850.py:

• Issue repro — a minimal instance using the opaque UNIT_STANDARD_HKD_... id resolves to currency == "HKD" while unit_ref stays opaque.
• Ground truth — the AAPL 10-K fixture resolves USD-denominated facts (and usdPerShare) to "USD".
• Silence check — share-count facts have currency of None, not a fabricated code.

All three are red without the change. The existing facts/unit suites (test_xbrl_unit_pointintime.py, test_xbrl_facts.py, …) show no new failures (the remaining failures are pre-existing network-dependent tests). ruff check/format clean on the changed code.