Reject extra array elements for fixed-length types (text and Value)

[yuxi-liu-wired]

fix: reject extra array elements for fixed-length types

A tuple, array, tuple struct or struct read from an array asks the SeqAccess for exactly as many elements as it has, then stops. The elements it left unread stayed on the tape and were read as the values after the array, so the data silently shifted:

from_slice::<((u8, u8), u8)>("[[1,2,3],4]")          Ok(((1, 2), 3))
from_slice::<(u8, (u8,), u8)>("[1,[2,99,98],3]")     Ok((1, (2,), 99))
from_slice::<Vec<(u8, u8)>>("[[1,2,[9,9]],[4,5]]")   Ok([(1, 2), (9, 9)])

In other shapes it failed with unrelated errors (ExpectedArray, ExpectedString). serde_json and sonic-rs reject all of these.

After the visitor returns, the remaining element count must be 0. Otherwise it is an invalid_length error, as in serde::de::value::SeqDeserializer::end and serde_json's visit_array. The same check covers objects, where a visitor stops reading members early.

The Value deserializers got the same check in #488.

Tests:

• fixed_length_sequences_reject_extra_elements
• tuple_variants_reject_extra_elements (text and every Value, as asked in review)

Both fail on main and pass with the change. The full test suite passes, and cargo fmt --check and clippy are clean.

Found with a differential fuzzer comparing serde_json, simd-json, sonic-rs and jiter.

[Licenser]

just some testy stuff

[Licenser]

will need a re-base due to conflicts, otherwise 🚀

[yuxi-liu-wired]

Rebased on main. Since #488 already brought in the Value side of this (visit_array in the owned and borrowed Value deserializers, with value_fixed_length_sequences_reject_extra_elements), I dropped that commit. What's left is the text deserializer fix and its two tests, fixed_length_sequences_reject_extra_elements and tuple_variants_reject_extra_elements. Both fail on main and pass here. The full test suite, fmt and clippy are clean.