Document six missing aggregation pipeline stages

[GuanzhouSong]

Continues @richardsimmonds' work in #47, which this supersedes. His original commit is preserved in this branch's history; the reference gap it identified was real, and six stages that were missing are documented here.

What this adds

$currentOp, $replaceRoot, $search, $setWindowFields, and $unionWith — five stages with no reference page today.

What changed from #47

Four files dropped.$graphLookup, $limit, $project, and $vectorSearch were documented in #59 while #47 was open, and were add/add conflicts against pages that now exist.

$search rewritten. The page documented Atlas-style full-text search over a text index. DocumentDB's $search is a vector search stage — SearchOperatorsList[] holds exactly cosmosSearch and knnBeta (deprecated), and there is no text operator in the search sources, so the original example failed with Unrecognized $search option: text rather than returning nothing. Rewritten against the stage's parsing: both operators, the options carried alongside them (index, count, returnStoredSource), and the operator spec (path, vector, k, filter, exact, oversampling, score), including which fields knnBeta rejects and which need server-side pre-filtering. count is noted as having no observable effect, since the engine parses it but does not yet emit the metadata.

$setWindowFields corrected. The syntax template put documents, range, and unit in one window block, a combination EnsureValidWindowSpec rejects; it now shows them as the alternatives they are, with both parse errors quoted. sortBy was described as plainly optional — a range window requires a single ascending sort field (so the -1 the table endorsed cannot be used with one), a bounded documents window requires one, and several operators require their own. Added as a table with the error each case raises.

$count dropped. It duplicated accumulators/$count.md, which already documents the stage including the same $unwind example, and the two disagreed on collection size. Removing the new page keeps the stage documented and ends the duplicate title. Filing $count under aggregation/ instead would be a move of the existing page, and belongs in its own change.

$replaceRoot examples made self-consistent. Both ended in { $limit: 2 } while showing a single document. The count was the error, so the limit now matches the output — true by inspection rather than dependent on collection size, which is what made the originals wrong.

Verification

Claims on the rewritten and corrected pages were derived from the engine source rather than from MongoDB's documentation, which is how the original divergences arose — everything wrong in #47 was accurate for MongoDB or Atlas. $currentOp was checked and left alone: its five options are real and its first-stage and admin-database requirements match the handler.

[guanzhousongmicrosoft]

Approving. The six pages fill a real gap, and the claims on the rewritten pages are derived from the engine source rather than from MongoDB's documentation - which is what the original divergences came from. Credit to @richardsimmonds for the original work in #47.