refactor: decompose archive read/write into composable building blocks

Delete archive_context.rs and archive_ops.rs (1200+ lines of duplicated
logic). Replace with four focused modules:

1. open_archive() - opens a file, detects compression, returns raw bytes
2. read_archive() - parses bytes into validated observations
3. CompressionWriter - writes bytes with any compression format
4. WriteStrategy - given a list of files, determines input archive,
 output archive, output format, and which of four write modes to use:
  - Create: new archive, no input
  - Append: uncompressed input, seek to end
  - AtomicSwap: compressed input, rewrite via temp file
  - CopyOnWrite: different input/output paths, transcode between formats

Previously you could not specify output format. Appending always
preserved the input format, creating compressed archives didn't work.
Now all four cases work with any supported compression format.

Atomic swap now writes to temp file, then renames. Crash-safe.

Trade-off: This approach prioritizes code clarity over syscall efficiency.
  The archive file may be opened and read multiple times during a single
  operation (once for format detection, once for reading state, once for
  copying content). A more optimized implementation could reuse file
  handles, but the current approach makes each step's purpose obvious.

Cargo.toml

@@ -1,8 +1,9 @@
 [package]
 name = "json-archive"
-version = "0.99.0"
+version = "0.99.1"
 edition = "2021"
 authors = ["Karl <marxism@peoplesgrocers.com>", "nobody <nobody@localhost>"]
+homepage = "https://peoplesgrocers.com/code/oss/json-archive"
 repository = "https://peoplesgrocers.com/code/oss/json-archive"
 license = "AGPL-3.0"
 description = "CLI tool for tracking JSON file changes over time using delta-based archives"