fix: preserve absent components across Serialization roundtrip

[theogf]

URI distinguishes an absent component from an empty one using a sentinel SubString (absent) identified by object identity (===). The default Serialization machinery round-trips field values but not object identity, so a deserialized URI's components were no longer === to absent, corrupting URIs built from components (e.g. a file: URI gaining a bogus empty authority). Serialize/deserialize through the URI string instead.

Fixes #75

Written by Claude Opus and manually checked.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.56%. Comparing base (859a10c) to head (a0e095e).

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #76      +/-   ##
==========================================
+ Coverage   86.91%   87.56%   +0.65%     
==========================================
  Files           5        5              
  Lines         382      402      +20     
==========================================
+ Hits          332      352      +20     
  Misses         50       50              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[quinnj]

Thanks! One blocker: this breaks data serialized by existing releases. The old format has eight fields, but the new reader consumes only one, which misaligns the stream. For example:

# Written by the current release, then read by this PR:
(URI(; scheme="file", path="some/relative/path.pdf"), 42)

# Becomes:
(URI(""), "file")

Could we add a format marker, a backward-compatible reader, and a regression test with a trailing value?

[quinnj]

Thanks, the stream-alignment issue is fixed. One legacy case remains: empty components are always converted to absent. A v1.7-serialized URI("http://example.com?") loses the ? after a later component update. Could we reparse a nonempty cached uri after consuming the legacy fields and add a regression test for this case? Cacheless legacy values can keep the documented best-effort fallback.