Wren 0.6.26

Wren 0.6.26 runs pages that build large documents again much faster, and reads pages written in legacy encodings right.

This is still a beta, under development.

Pages that rebuild their DOM are much faster: elements alike share their selector matching, document and window stay on the interpreter's fast path, a wrapper is one allocation, string arguments are read in place and a collection unpins its nodes in one call (a page that rebuilds 10 000 nodes every frame, measured on one machine: about 85 -> 142 frames a second since 0.6.25).
Legacy encodings: a page in Shift_JIS, EUC-KR, Big5, GBK or windows-1252 keeps its encoding for its links and forms (a search form no longer sends mojibake), document.characterSet says it, level-2 kanji decode, and documents made by a script keep their own encoding and URL.
getComputedStyle gives the used insets (top, right, bottom, left in pixels) of positioned elements: what tooltip and popover libraries read.
<dialog>: beforetoggle and toggle events, closedBy, and a modal dialog leaves the top layer when it is taken out of the document.
document.createProcessingInstruction, and an error of a method of another frame's node is of that frame.
Text on several threads waits less (the graphics toolkit's atlas and word cache locks).

#Wren 0.6.26