Reduce duplicative instantiation of logic in SeqAccess and MapAccess

[dtolnay]

Measured in https://github.com/serde-rs/json-benchmark using cargo llvm-lines --bin json-benchmark --no-default-features --features parse-struct,lib-serde,all-files --filter 'next_element_seed|next_key_seed', this PR reduces LLVM lines by 53% (compared to 37% for #1102). This is the part of the logic that needs to be instantiated separately per seq/map type, as opposed to once per Read type.

Across the whole json-benchmark program, this PR decreases total LLVM lines by 8.8% (compared to 5.6% for #1102) and decreases function instantiations by 3.4%.

Benchmarked using cargo build --release --bin json-benchmark --no-default-features --features parse-struct,lib-serde,all-files, there is no statistically significant performance difference at runtime. (But I notice some opportunities to improve performance of the code in these functions. Leaving for a future PR.)

Also benchmarked using the following microbenchmark that targets this code specifically, which likewise shows no regression.

#![feature(test)] extern crate test; #[bench] fn bench(t: &mut test::Bencher) { use serde::de::{Deserialize, Deserializer, SeqAccess, Visitor}; use std::fmt; struct Arrays; impl<'de> Deserialize<'de> for Arrays { fn deserialize<D>(deserializer: D) -> Result<Self, D::Error> where D: Deserializer<'de>, { deserializer.deserialize_any(Arrays) } } impl<'de> Visitor<'de> for Arrays { type Value = Arrays; fn expecting(&self, formatter: &mut fmt::Formatter) -> fmt::Result { formatter.write_str("arrays") } fn visit_u64<E>(self, _: u64) -> Result<Self::Value, E> { Ok(Arrays) } fn visit_seq<A>(self, mut seq: A) -> Result<Self::Value, A::Error> where A: SeqAccess<'de>, { while seq.next_element::<Arrays>()?.is_some() {} Ok(Arrays) } } let mut json = "0".to_owned(); for _ in 0..16 { json = format!("[{json},{json}]"); } t.bytes = json.len() as u64; t.iter(|| serde_json::from_reader::<_, Arrays>(&mut json.as_bytes())); }

Closes #1102.