fix: detect upper-accented+currency mojibake at string start (fixes #222)

[gaoflow]

The badness heuristic in is_bad() misses UTF-8 mojibake when an upper-accented letter followed by a currency symbol appears at the very beginning of a string. An existing pattern (\s [{upper_accented}] [{currency}]) already catches this when preceded by whitespace, but the start-of-string case was missing.

For example, fix_encoding("Ã¥klagarmyndighets") returns "Ã¥klagarmyndighets" unchanged instead of correctly decoding to "åklagarmyndighets".

Root cause: No BADNESS_RE pattern matches [{upper_accented}][{currency}] at position 0 without a preceding space or lowercase letter.

Fix: Add ^[{upper_accented}][{currency}]\w to detect this sequence at string start. The trailing \w is required so the pattern does not match the isolated 2-character substring that decode_inconsistent_utf8 passes to is_bad(), preventing false positives on ambiguous embedded sequences (like the "Dråber" negative test case from issue #202).

Changes:

• ftfy/badness.py: Add new BADNESS_RE alternation (4 lines)
• tests/test-cases/synthetic.json: Add test case for the reported bug

[sebollin]

I applied this on top of 74dd045 and it does what it says:

• fix_encoding("Ã¥klagarmyndighets") → åklagarmyndighets
• the whitespace case still works
• suite unchanged at 351 passed, 10 xfailed

The trailing \w is load-bearing, as the description says. Without it the alternative also matches the two-character fragment that decode_inconsistent_utf8 looks at in isolation — took me a minute to see why it was there.

One thing for whoever reviews this: that rule is the only alternative pairing upper_accented with currency, so (?:^|\s) in place of \s would cover both positions in a single alternative instead of two, if that shape is preferred.