fix: accept every PDF standard and fail on invalid combinations

[msallin]

Problem

Two defects in how pdfStandards is handled, both ending in a document that does not conform to what was requested.

PDF/UA-1 could not be requested. The accepted names were a hand-written match in compile_internal. It listed the PDF versions and the PDF/A levels but omitted ua-1, so the accessibility standard was rejected outright even though typst-pdf 0.15.1 supports it. Any standard added by a later Typst release would need the table updated by hand.

Invalid combinations were silently downgraded. export_pdf called PdfStandards::new(standards).unwrap_or_default(). When a combination is contradictory the validation error was discarded and export fell back to the default configuration, producing an ordinary PDF while reporting success. A pipeline could believe it was writing PDF/A until an archive validator disagreed.

Running the new tests against develop:

[Test Failure] InvalidOperationException: Invalid PDF standard: ua-1
[Test Failure] AssertionException: Expected exception message to contain "PDF/A"
[Test Failure] AssertionException: Expected exception message to contain "same time"
[Test Failure] AssertionException: Expected to throw InvalidOperationException

The last is the silent downgrade: ["a-1b", "v-2.0"] contradicts itself, and the export succeeded anyway.

Change

Standard names are parsed through PdfStandard's own serde representation instead of a table, so every variant Typst defines is accepted and future ones need no change here. PdfStandard is #[non_exhaustive], so an exhaustive match was not possible from outside the crate in any case. The v- prefix on plain PDF versions is kept as an alias because the README documents v-1.7.

export_pdf propagates the error from PdfStandards::new instead of discarding it, including the hints Typst attaches, which name the PDF versions each standard allows. The wording for an unknown name is unchanged, so InvalidPdfStandardThrowsException still passes.

Tests

Eight Rust unit tests over the parser cover every alias form, ua-1, case-insensitivity, blank entries and unknown names. Six tests in Tests.cs cover the behaviour end to end: ua-1 marks the output with pdfuaid:part, a-2b marks it with pdfaid:part where a plain export does not, and each contradictory combination throws with the upstream message.

cargo test --release   11 passed
dotnet test            51 passed

Note for reviewers: PDF/A requires the document to carry a date, and the exporter deliberately sets no timestamp, so the conformance tests set one in the template. This is worth documenting for users adopting PDF/A.

[msallin]

Superseded by evolvedlight#38; the fix belongs upstream.