[RFE] Support compression/decompression for large diagnostic KVP payloads (e.g. dmesg)

### Summary

Add support for capturing large diagnostic payloads (e.g. `dmesg` / boot logs), compressing them, and writing them to the Hyper-V KVP guest pool and decompressing them on read. Cloud-init already does this (it compresses and chunks `dmesg` into KVPs), and it's proven useful for boot/performance analysis.

### Current state

- `DiagnosticsKvp` (added in #313) already chunks large values across KVP records at UTF-8 boundaries and reassembles them on read but values are stored as plain UTF-8, uncompressed.
- The store validates values as UTF-8 and rejects NUL bytes, so raw compressed bytes can't be stored directly; a text-safe encoding (e.g. base64) is required on top of compression.

### Proposal

1. Write: `compress → base64-encode → chunk`, reusing the existing chunking + `append_multiple`.
2. Read: `reassemble (existing) → base64-decode → decompress` back to the original payload.
3. `diag` CLI: transparently decompress compressed records when displaying (new `compressed` record kind or marker), with a `--raw` option to show the encoded form.
4. Provide the compress/encode/chunk primitive in `libazureinit-kvp`; leave capture of a specific source (e.g. `dmesg`) to the azure-init binary.

### Related

- Diagnostics layer (chunking/reassembly this builds on): #313

Start by reading DiagnosticsKvp and the existing chunking and reassembly path from #313, then inspect the libazureinit-kvp primitive and the diag CLI entry point. Done means large payloads round-trip through compression, text-safe encoding, and chunking, while diag displays decompressed records by default and supports the proposed raw form.

Written by the indexing model from the issue text.