Exclude prose/data from language stats; documentation exclusion; programming-only dominant language

[nielsbosma]

Implements the three language-statistics accuracy suggestions for github-linguist parity. All three parts are done (none deferred).

Part 1 — Exclude prose + data from language statistics

LanguageAggregator.Aggregate now skips LanguageType.Prose and LanguageType.Data entirely (rows and denominator), so percent is the share of programming + markup bytes only. XML-doc updated. (Ref: language-aggregator-exclude-prose-data.md)

Part 2 — Documentation/example exclusion

• New embedded data file src/Ivy.StackAnalyzer/data/documentation.yml (seeded from linguist documentation.yml: examples/samples/demos/docs/documentation/changelog/README/license path patterns).
• New DataStore.LoadDocumentation (resource suffix .data.documentation.yml) mirroring LoadVendor, exposed as DataStore.DocumentationPatterns, with user-directory overlay support.
• New ScannedFile.IsDocumentation, set in FileSystemScanner the same way as IsVendored but WITHOUT pruning the directory from the walk (example/doc dirs still surface as [auxiliary] components; ComponentDetector still reads the unfiltered list).
• Aggregate guard extended with || f.File.IsDocumentation.
  (Ref: language-examples-overcount.md)

Part 3 — Programming-only dominant language

Added LanguageAggregator.DominantProgrammingLanguage, restricting the dominant/headline language pick to type == Programming (linguist parity), so a data (JSON) or prose (Text) language can never out-rank the real code language. After Part 1 the per-component Languages list (consumed for the hash language slot) is already free of data/prose; this helper makes the rule explicit and also excludes markup. (Ref: dominant-language-includes-data-prose.md)

Tests

• dotnet test: 97 passed / 0 failed (95 baseline + 2 new aggregator tests).
• Unit tests updated in LanguageAggregatorTests.cs: rewrote Aggregate_excludes_* to assert prose/data exclusion + re-percent; added Aggregate_excludes_documentation_files and DominantProgrammingLanguage_ignores_markup_data_and_prose.
• Snapshots updated (all 4): go-service (drops Go Module data row), numpy-style-lib (drops TOML), django-spa (drops JSON/Markdown/Pip Requirements), next-dotnet-monorepo (drops XML/JSON/YAML/Markdown). Percentages re-computed over programming+markup; docs-only auxiliary subtrees now show languages: [].

Note: no test fixture contains an example/docs subtree of real code, so Part 2's path exclusion is covered by the dedicated unit test rather than a snapshot.