Summarize a model from an already-parsed structure

[ww-mw]

A host opening a model parses it twice: once itself, to build the model rows, and once inside summarizeFiles, to build the usage summary over the same bytes. The second parse produces nothing the caller was not already holding.

summarizeParsedModel takes the parse instead of the bytes. It is the model branch of summarizeFiles with the parse lifted out, and summarizeFiles' own model branch now routes through it — so the two ways to summarize a model are one implementation rather than two that happen to agree.

It answers with a whole FileSummaries holding just that one model rather than a bare ModelSummary, so it is exactly what summarizeFiles([{srcId, filename, bytes}]) returns for a model, and mergeFileSummaries folds it in beside the dictionaries and MAT-files a host only ever reads as bytes.

Purely additive — no existing signature or behaviour changes. The tests pin the equality BETWEEN the two paths per model fixture: block parameters, a real model workspace, mask workspaces, file links with no blocks, and a model that parses and defines nothing (where both paths owe the same empty-ish summary rather than a throw). Bytes the parser refuses have no parsed path at all — the caller's own parseModel threw and has nothing to hand over — and that is pinned as the one case the two cannot be compared.

Version bumped to 1.20.0 on the branch so the release tag can match, per the pattern in 684e36c.

Consumer: data-explorer-vscode's shared-source-cache work, which calls this to build the Usage graph from the parse it already has.