Add "Choosing a collection" section to README

[marius-bughiu]

Summary

• Adds a new Choosing a collection section to README.md, between Quick start and Benchmarks. It contains a decision table mapping common workloads (int-keyed, long-keyed, Guid / string / other-keyed dictionaries, the two set shapes) onto the right Celerity type, a short note on picking a hasher (deferring to the existing matrix in docs/api/hashing.md), and a candid list of cases where Celerity is not the right answer today: concurrent access, mutable IDictionary<,> consumers, and FrozenDictionary-style build-once string-keyed lookups (the Celerity equivalent is on the 1.2.0 roadmap, #62).
• Adds a corresponding [Unreleased] entry to CHANGELOG.md.

This implements the "Document when to use which collection" sub-task from issue #15. Issue #15 has several other sub-tasks still open (perf-tuning guide, migration guide, troubleshooting, FAQ), so the issue itself stays open.

Doc-only change — no source under src/Celerity/ was touched.

Test plan

• cd src && dotnet build — 0 errors (warnings are pre-existing xUnit lint nits, unchanged by this PR).
• Render the README on GitHub and confirm the new table sits cleanly between Quick start and Benchmarks.
• Skim the section as a first-time reader and confirm the workload-to-type mapping reads as expected.