Improvements in OTLP encoding logic required for internal logging safety

[jmacd]

Following #1735, here are a list of improvements we desire in the OTLP ProtoBuffer encoding object of crates/pdata/src/otlp/common.rs

1. Ability to encode into a limited-size buffer to avoid unbounded memory use
2. Ability to encode into a single pre-allocated &mut [u8]
3. Ability to safely truncate long strings appending a "..." or other notice
4. Ability to safely truncate a list of top-level fields (e.g., the attributes list of a log record)
5. Ability to automatically generate the dropped_attributes_count field of LogRecord, Span, Resource etc by reserving 5 bytes
6. Ability to choose (somehow) the number of bytes dedicated to placeholders; the current 4-byte approach is wasteful for small buffers

Ultimately with all of this done we will be able to safely log into pre-allocated buffers and obtain usefully truncated and/or formatted messages without allocating memory inside a tokio tracing event.

Part of #1736