fix(runtime): hydrate a tuple captured by a runtime expression

[AmeinEskinder]

Closes #254.

expr.md lists tuples in the shared vocabulary, and expr_field.rs:21 compiles pair.0 into JavaScript pair[0], so a tuple is an array on the browser side. Serde writes a tuple surrogate as one. hydrateSurrogate had no array case, so the array fell past the tag switch into the default arm and threw:

Error: Unknown surrogate type: undefined
 ❯ hydrateSurrogate src/surrogate/index.ts:73:12

Every expression capturing a tuple failed to hydrate, and scan walks the document in a single loop with no try/catch, so nothing after it hydrated either.

The change

Hydrate an array by hydrating its elements. The check sits before the tag switch because an array is an object with no t, and the result stays a real array so the generated pair[0] still indexes it.

That alone would trade a throw for a quieter disagreement, so both render paths follow the server too:

	server	browser before	browser after
$(pair) on (1.5, 2.5)	1.52.5	throws	1.52.5
:title=$(pair) on ("a", "b")	title="ab"	throws	title="ab"
tuple holding a None	element writes nothing	throws	element writes nothing

NodeViewParts for (T1, T2) writes its elements one after another with no separator, where String(array) would have given 1.5,2.5. AttributeValueViewParts for (T1, T2) is present when any element is, and skips the ones that are not, the way a None writes nothing — calling toAttributeValue() on a None would itself have thrown.

Verification

The strings below are what topcoat-runtime actually rendered for let pair = (1.5f64, 2.5f64), taken from a rendered view rather than written by hand:

$(pair.0)  ->  (() => { const [__external0] = [cx.hydrate([1.5,2.5])]; return __external0[0]; })()
$(pair)    ->  (() => { const [__external0] = [cx.hydrate([1.5,2.5])]; return __external0; })()

Running those exact strings through the patched runtime gives 1.5 and 1.52.5, matching the server's 1.5 and 1.52.5. Both threw before.

Eleven tests cover hydration, both render paths, and the wire format the fix depends on. I disabled each of the three new branches in turn and confirmed every test fails without it, with the failures being the divergences above (a,b for ab, 1.5,2.5 for 1.52.5, a,[object Object],b for a tuple holding a None).

The Rust-side tests pin the serialized shape ([1.5,2.5], [1.0,{"t":"Option","v":2.0},true], [[1.0,2.0],3.0]) so a change to tuple serialization can't silently break the browser half.

Checks

yarn build, yarn test (31 passed), cargo test -p topcoat-runtime, cargo clippy -p topcoat-runtime --all-targets (0 warnings), cargo fmt --check. dist/index.js is rebuilt.