xslt 1.6.13

This package has no dependencies.

Eleven fixes found by running XSpec's 284-suite corpus against the engine. On that corpus:
102 suites ran to completion before, 129 now; 505 assertions passed before, 1077 now. No suite
lost ground at any measured step.

### fn:transform dropped parameters, and mis-identified namespaced names

`template-params` and `tunnel-params` never reached an initial MODE: `TransformAsync` passed its
with-param list and `TransformRawAsync` passed an empty one at all three of its call sites.
XSpec compiles a scenario-level `x:param` under `x:context/@mode` into `template-params`, so
whole suites ran with their parameters missing and terminated on the first assertion that read
one.

Separately, every name in an `fn:transform` option map arrives from `fn:QName()` at runtime,
carrying an interned id the parser never assigned. Namespaced names missed every lookup - and a
namespaced `initial-mode` collapsed to no namespace and ran a DIFFERENT mode that happened to
share a local name.

`initial-match-selection` was also round-tripped through an XPath string, so a sequence became
the CLR text `System.Object[]`. No expression can denote an arbitrary existing node, so that
conversion could never work; the value is carried now.

### xsl:attribute with a body escaped its content twice

The body form runs its content into the ordinary output buffer, which escapes for element
content, and the finished value was escaped again. `&` came back as `&`. The select and
AVT forms were always correct, so the divergence was between forms.

Nothing errors - the attribute is well-formed, just wrong - so it only surfaces when something
reads the value back. XSpec's compiler builds a `select` attribute this way around an XPath
containing `=>`, and the generated stylesheet then failed to parse. Worth 148 assertions.

### A forward-declared global with a NAMESPACED name was evaluated too early

Dependencies were collected by testing a variable reference for membership in the set of
declared names. The reference comes from the XPath parser and the declaration from the
stylesheet parser, and their interned ids can differ - so for a namespaced name the lookup
missed and the dependency was never recorded.

Nothing errored. The topological sort simply did not order the pair, and a global declared
before the ones it selects from read unbound values. It surfaced far from the cause as node
kinds collapsing: `comment()` and `element()` came back as documents, `attribute()` as an empty
string, `namespace-node()` as text.

### Other correctness fixes

- Supplied parameters overwrote global `xsl:variable`s. XSLT 3.0 §9.5: a value supplied for a
 name declared as a variable is not a parameter and must have no effect.
- The internal text marker leaked into template matching, type coercion and node-typed
 variables - three places that reasonably assume a real node.
- A global `as="text()"` variable took an id from the node store without registering it.
- Union-pattern siblings forked their group identity, so two branches of ONE rule stopped
 recognising each other and matching both raised XTDE0540 (spec bug 30402).
- `as="processing-instruction(name)"` bound a document; the named spelling had no branch and
 fell to a silent `item()` default.
- `SetInitialTemplate` mis-parsed an EQName, splitting `Q{http://x}main` at the URI's colon.

### Errors that named the code and not the problem

XTDE0555, XTTE0570, XTDE0540 and XPDY0002 reported the error code and the expected type - the
two things the author already knows - and not the actual node, mode, value or conflicting rules.
XTTE0570 was emitted from nine sites with identical text. A CLR type name was also reaching
users as `match="PhoenixmlDb.Xslt.Ast.DotPattern"`.

Three clusters were then solved in a single run each by reading the improved message.

   Xslt.Tests 1416 passed, 0 failed