refactor(github): attribute commits by repo language byte share

Each commit now contributes a full "vote" partitioned across the repo's
languages proportional to linguist byte counts, instead of crediting only
the primary language. A 60% Go / 40% Python repo adds 0.6 to Go and 0.4
to Python per commit.

- RepoInfo gains []LangEdge capturing the full byte breakdown already
  returned by profileQuery.
- FetchProductive distributes each commit via a fixed-point scaleFactor
  (int64 preserved, percentages unchanged in the card).
- Fallback to primary language only when linguist reports zero bytes
  (empty repo).

Caveat: linguist excludes prose languages (Markdown, AsciiDoc, reST) from
its byte output, so Markdown-heavy repos still skew toward the detected
code fraction. Fixing that case requires per-commit file classification
via REST /commits/{sha} + go-enry — tracked as future work.

internal/github/model.go

@@ -56,6 +56,17 @@ type RepoInfo struct {
 	Name            string
 	PrimaryLanguage string
 	PrimaryColor    string
+	// Languages is the repo's language byte breakdown as reported by GitHub
+	// linguist. Used to attribute each commit fractionally across all
+	// languages the repo contains, rather than crediting only the primary.
+	Languages []LangEdge
+}
+
+// LangEdge is a single entry in a repo's language-bytes breakdown.
+type LangEdge struct {
+	Name  string
+	Color string
+	Bytes int64
 }
 
 // repoNode is the GraphQL shape of one repository node; kept here because