Read a URL encoded CSV in the #csv= fragment

[aaronj1335]

#csv= took only base64url bytes, so the small hand-written link -- the
one ?csv= already takes -- had to go through a gzip pipeline before it
could stay out of the request. It now takes percent-encoded CSV text too.

Which encoding a link uses is read off the payload rather than declared:
base64url spells everything in A-Z a-z 0-9 - _ =, and a CSV needs the
comma between its date column and a value column, so a payload holding a
character outside that alphabet is text and never a mangled base64url
string. A base64url payload run through encodeURIComponent arrives with
its padding as %3D, which is why the percent decoding happens first.

URLSearchParams no longer reads the fragment. It decodes what it
returns, which would decode a percent-encoded CSV here and again in
decodeCSVFragment -- a cell holding the literal %0A would come back as
a row break -- and it form-decodes, reading a + in a cell as a space.
The raw value is handed over instead, and decoded exactly once.

The percent decoding works escape by escape rather than over the whole
string, because decodeURIComponent is all-or-nothing: one cell holding a
bare % would otherwise cost the newlines in every row.

Co-Authored-By: Claude Opus 5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_017hKewm4Qv8pVHVYDfBYbHg