xquery4 2.4.0

This package has no dependencies.

Takes **PhoenixmlDb.Core 2.0.0** (unchanged).

### Fixed: typed results the conformance harness used to hide (#83, #85)

Before #49 the QT3 harness counted any `assert-type` it didn't recognise as a pass. Once it
checked for real, six engine defects showed up that had passed by default:

| | was | now |
|---|---|---|
| `fn:namespace-uri` | `xs:string` | `xs:anyURI` |
| `fn:prefix-from-QName` | `xs:string` | `xs:NCName` |
| `fn:default-language` | `xs:string` | `xs:language` |
| `map:put` with an equal key of another type | kept the old key | replaces key and value, same position |
| `xs:short(256) cast as xs:numeric` | lost `xs:short` | keeps it |
| `schema-element(unbound:x)` | `XPST0008` | `XPST0081` |

The new types reach code that had only seen plain strings, so these were fixed at the root: one
rule for `xs:string` arguments means `string-to-codepoints`, `codepoint-equal` and
`encode-for-uri` accept `xs:anyURI` and string subtypes, and string constructors accept a
string-subtype source.

### Behaviour changes — read before upgrading

- **`fn:namespace-uri` returns `xs:anyURI`.** Code that tests its result `instance of xs:string`
 gets a different answer.
- **`fn:distinct-values` treats an `xs:anyURI` and an equal `xs:string` as one value**, matching
 this engine's own `eq` and XPath 4.0. This reverses an earlier choice whose spec citation
 didn't hold up.
- **Default output renders `xs:anyURI`, `xs:untypedAtomic` and string subtypes bare**, the same
 way a plain `xs:string` already rendered. Without this, every `namespace-uri()` result
 printed through the facade would have gained quotes. A query that declares
 `output:method "adaptive"` still gets the quoted W3C form.

### Conformance

**W3C QT3: 29,895 / 31,379 (95.27%)**, measured on the release commit. The denominator is
unchanged. #85's per-case diff shows 14 cases newly passing and none newly failing: that fully
recovers the nine the harness correction exposed, plus five more.

The ratchet in `scripts/conformance-baseline.tsv` is raised to this measurement. It hadn't been
raised since before 2.2.0, so part of the raise (+91 across 19 sets) is earlier work that was
never recorded, for example 2.2.0's `xs:IDREFS` cast targets. It is not all this release.

### `xquery` CLI (ships on `cli-v2.4.0`)

- **Output methods: `html` and `xhtml` are supported, and an unrecognised method is now an error**
 (#87). Before this, `-o html`, `declare option output:method "html"` and any typo all silently
 fell back to adaptive output with exit code 0. Now `-o` accepts `adaptive`, `xml`, `html`,
 `xhtml`, `text` and `json`; any other value exits with code 1, and an unknown method in the
 prolog raises `SEPM0016`. **Scripts that relied on the silent fallback will now fail visibly.**
- **Output is UTF-8 on every platform** (#80; no earlier CLI release carried it). On
 Windows it used to follow the console's legacy code page, which corrupted non-ASCII characters
 in redirected output.
- Runs on PhoenixmlDb.Xslt 2.4.0 for `fn:transform`, so `source-location` and raw delivery of
 constructed nodes both work from a query.

### Upgrading

`PhoenixmlDb.Xslt` 2.4.0 needs its own fix for this change (xslt#188, promoting `xs:anyURI` to
`xs:string` for stylesheet-function parameters). Without it, a stylesheet function receiving
`namespace-uri()` fails. Take the two packages together; the lockstep pins enforce that.

**2.3.0 carried no library change** (it was byte-identical to 2.2.0, a lockstep bump that was
published and then superseded). Moving from 2.2.0 or earlier straight to 2.4.0 is the intended
path.