fix(autocomplete): compare whole completions when removing duplicates

[boogie]

Problem

FilteredList.setFilter removes duplicates by comparing each completion with the one immediately before it. That is only correct while equal completions stay adjacent, and they do not: the list is sorted by score first, so a duplicate scoring differently from its twin ends up separated by another completion and both reach the popup.

Two completers offering the same word with different scores is enough:

[{value: "foo", score: 5}, {value: "bar", score: 3}, {value: "foo", score: 1}]
// popup shows: foo, bar, foo

The comparison is too coarse in the other direction as well. The key is snippet || caption || value, so snippets are compared by their body alone, and two snippets expanding to the same text collapse into one however differently they are captioned. Ace's own html mode ships fourteen doc snippets that expand to only eight distinct bodies — three of them are dropped from the popup today, and the survivor of each pair is whichever sorts first alphabetically.

Change

Compare what a completion reads as in the popup together with what it inserts, and track every completion kept rather than only the previous one.

	before	after
same value, separated by a different score	both shown	merged
same value, adjacent	merged	merged
distinct captions, same snippet body	merged	both shown
html mode, doc prefix	11 of 14 shown	14 of 14 shown

Completions that look the same in the popup and insert the same text are still merged, which is what keeps the keyword and text completers from offering the same word twice. Of a set of duplicates the first one after sorting survives, which is the highest scored — the same one that survived before.

The Set replaces a single variable in a function that already sorts the same array immediately above, so the added cost is not measurable against what setFilter does per keystroke.

Test

test: duplicate completions are removed regardless of their position covers both directions at once, and asserts the surviving duplicate is the higher scored one. Against the current code it fails with:

'["foo","bar","foo","snippet a"]' == '["foo","bar","snippet a","snippet b"]'

— foo twice, and snippet b missing.

Open kitchen-sink @ 2a773af408f8dfa2c8f61df011b845662b2c4e02