Fix: URI->new no longer dies on malformed geo: URIs

[oalders]

Closes #201

Changes

• URI::geo::_init probes the latitude under do { local $@; eval { ... } } and only does the pole step when the path parses. A malformed geo: string now builds a URI::geo object, like every other scheme, and the caller's $@ is left alone.
• The accessors (latitude, longitude, location, field, ...) still croak Badly formed geo uri, so that check now happens when a value is read, not in the constructor.
• POD DIAGNOSTICS: removed the duplicated Badly formed geo uri item and said where it is raised.
• POD notes that up to and including 5.37 URI->new died on such input, and that 5.38 changes this.
• Changes entry under {{$NEXT}}.

Behaviour change: t/geo_basic.t used to assert that URI->new("geo:1") dies. That came with the original port of Andy Armstrong's standalone URI::geo (9d9b36c, #141). Inside URI it is the outlier: URI::urn::_init falls back on a bad NID rather than croaking. That test is replaced with the new behaviour. Code that used eval { URI->new(...) } to validate geo: strings should call an accessor instead.

Testing

• New regression test in t/geo_basic.t: URI->new("geo:1") returns a URI::geo, leaves a sentinel $@ alone, and ->latitude croaks. It fails on the old geo.pm (dies at line 52) and passes with the fix.
• prove -lr t: 63 files, 1038 tests, PASS. precious lint is clean.
• /code-review-intense-flow (general, security, test-value): clean after one fix round (local $@) and again after the POD version note. Local Codex review: clean.

Declined findings

• Extra tests for geo:90,x, geo:a,b, and the location accessor: these take the same _parse failure path as geo:1.
• Test for geo:91,0 croaking Latitude out of range: the premise is false. _parse does no range check, so latitude returns 91 (checked).
• An is_valid helper and a fuzz test: nothing in the issue or in any caller needs them (YAGNI).
• An extra POD sentence about validation: the DIAGNOSTICS entry already says URI->new does not raise it.

🤖 Generated with Claude Code · Opus 5.5

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.14%. Comparing base (1f9410f) to head (e5edf89).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #202      +/-   ##
==========================================
+ Coverage   87.13%   87.14%   +0.01%     
==========================================
  Files          64       64              
  Lines        2099     2101       +2     
  Branches      512      512              
==========================================
+ Hits         1829     1831       +2     
  Misses        110      110              
  Partials      160      160              

☔ 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.

[esabol]

Why would you not want malformed URIs to die in new()?

[oalders]

@esabol no other package in this URI distribution behaves this way, so it's an outlier. This change should bring it in line with the other schemes.