chore: prepare 0.10.3

[cardmagic]

Bumps to 0.10.3 and records a QA pass over the whole gate.

Why 0.10.3

Two fixes and a documentation gap since 0.10.2, all reported from production use, none of them API changes:

• reminder uniqueness was undocumented, and an actor arming one alarm per queued item silently kept only the last
• solid_objects.reminder.replaced makes that replacement visible
• TestHelper#reset_actors! left seven tables behind wherever the database cascade is not enforced

A patch bump is right: nothing added to the public API except one instrumentation event, and no behaviour changed for existing callers.

QA pass

Everything below was run by hand on this branch.

Check	Result
bundle exec rake (SQLite)	390 runs, 1303 assertions, 0 failures, 14 skips
mysql2 (docker MySQL 8)	390 runs, 1263 assertions, 0 failures, 23 skips
Trilogy (docker MySQL 8)	390 runs, 1263 assertions, 0 failures, 23 skips
PostgreSQL 17.6	390 runs, 1280 assertions, 0 failures, 15 skips
Redis wake-up (docker, 6380)	10 runs, 0 failures, 0 skips
PostgreSQL wake-up	10 runs, 0 failures, 0 skips
npm test / npm run test:browser	26 pass / 13 pass
Standard, RuboCop, RBS, Steep, Brakeman	clean
gem build + contents	109 Ruby and JS files, no test/, no node_modules
Doctor against a real app	6 checks, healthy, reports schema matches the 0.10.3 runtime
CLI denial path	prints the policy and the setting, exits 1
benchmark/query_count.rb	26 / 49 / 75, matching the documented figures
ruby -c on shipped examples	Syntax OK

Both wake-up suites were run with their services configured rather than left to skip, since a skip and a pass look identical in the summary line. Both MySQL clients were run, since an adapter name, a connection collation, and an error code name all differ between them.

Findings

Nothing new. The three previous release QA passes each turned up something; this one did not, which is what I would expect now that the CLI, benchmark, and adapter gaps found in 0.10.0 and 0.10.1 are closed and the load contract test guards the class of bug behind 0.10.2.

One documentation correction, from re-reading the roadmap against the code: the "opt-in Minitest helper" entry now records that the helper clears actor-owned tables itself rather than relying on a cascade a host application may not enforce. That is the behaviour shipped in #31, and the entry read as if the helper simply worked.

The query counts are worth calling out because they are deterministic rather than machine-dependent: 26 for a message turn, 49 for the caller of a synchronous call, 75 combined. Unchanged since they were re-measured for 0.10.0, so nothing in the last three releases has moved the per-call database cost.

Ready to tag once merged.

[greptile-apps]

Greptile Summary

This PR prepares the 0.10.3 release by synchronizing package metadata and publishing the accumulated changelog. It also clarifies that the Minitest helper directly clears actor-owned tables rather than depending on database cascades.

• Bumps SolidObjects::VERSION and lockfile metadata to 0.10.3.
• Promotes the unreleased changelog entries into the dated 0.10.3 release section.
• Corrects the roadmap description of test-helper cleanup behavior.

Confidence Score: 5/5

The PR appears safe to merge; its release metadata is internally consistent and its documentation accurately describes the existing implementation.

The changed files only update synchronized release metadata and clarify documented behavior that is supported by the helper implementation and its table-coverage tests.

Important Files Changed

Filename	Overview
lib/solid_objects/version.rb	Bumps the canonical gem version to 0.10.3, consistent with all other release metadata.
Gemfile.lock	Updates both local gem version entries to 0.10.3 without changing dependencies.
CHANGELOG.md	Converts the existing unreleased entries into the dated 0.10.3 release section.
docs/roadmap.md	Clarifies the already-shipped cascade-independent behavior of the Minitest reset helper.

_{Reviews (1): Last reviewed commit: "chore: prepare 0.10.3" | Re-trigger Greptile}