rust: add sans-io indexed reader

[james-rms]

Changelog

• Added: mcap::sans_io::SummaryReader and mcap::sans_io::IndexedReader structs, which can be used to a) parse out summary information and b) read messages efficiently using that summary information.

Docs

See cargo doc for usage info.

Description

Adds a sans-io indexed reading implementation, including conformance tests and benchmarks.

Benchmarks

The indexed reader is suspiciously faster than MessageStream by quite a bit. I suspect the major difference is that the indexed reader allows the caller to re-use the same buffer for every message yielded, rather than allocating a new one.

Raw cargo bench outputs (with changes removed because they compare to the previous run, which is confusing)

mcap_read_linear/MessageStream_1M_uncompressed
                        time:   [165.82 ms 167.64 ms 169.53 ms]
                        thrpt:  [5.8986 Melem/s 5.9653 Melem/s 6.0306 Melem/s]
Found 2 outliers among 10 measurements (20.00%)
  1 (10.00%) high mild
  1 (10.00%) high severe
mcap_read_linear/MessageStream_1M_lz4
                        time:   [203.02 ms 204.87 ms 207.63 ms]
                        thrpt:  [4.8163 Melem/s 4.8811 Melem/s 4.9255 Melem/s]
Found 1 outliers among 10 measurements (10.00%)
  1 (10.00%) high severe
mcap_read_linear/MessageStream_1M_zstd
                        time:   [213.31 ms 213.74 ms 214.26 ms]
                        thrpt:  [4.6672 Melem/s 4.6787 Melem/s 4.6881 Melem/s]
Found 1 outliers among 10 measurements (10.00%)
  1 (10.00%) high mild

mcap_read_indexed/IndexedReader_1M_uncompressed
                        time:   [42.935 ms 43.215 ms 43.655 ms]
                        thrpt:  [22.907 Melem/s 23.140 Melem/s 23.291 Melem/s]
Found 1 outliers among 10 measurements (10.00%)
  1 (10.00%) high mild
mcap_read_indexed/IndexedReader_1M_zstd
                        time:   [65.055 ms 65.616 ms 66.254 ms]
                        thrpt:  [15.093 Melem/s 15.240 Melem/s 15.372 Melem/s]
Found 2 outliers among 10 measurements (20.00%)
  1 (10.00%) low mild
  1 (10.00%) high severe
mcap_read_indexed/IndexedReader_1M_lz4
                        time:   [60.758 ms 60.977 ms 61.139 ms]
                        thrpt:  [16.356 Melem/s 16.400 Melem/s 16.459 Melem/s]

[Muon]

Why do we do this? Some extra explanation would be good.

[Muon]

Suggested change

	expected.sort_by(|a, b| a.1.cmp(&b.1));
	expected.reverse();
	expected.sort_by_key(|(_, log_time)| std::cmp::Reverse(log_time));

[Muon]

cloned() twice?