nostr: avoid a String per generic tag key in Filter

[JSKitty]

Description

Third of the follow-ups mentioned in #1425, after #1426.

Serializing a filter built each map key with format!("#{tag}"), allocating a String per generic tag query. The key is always # followed by a single ASCII letter, so it fits in a two-byte buffer on the stack.

Deserializing had the mirror problem: every key in the map was read into an owned String just to inspect its first two characters. Reading it through a small key type that looks at the borrowed &str and yields the tag directly avoids that.

Measured on an Apple M4, filter with three generic tag queries (178 bytes):

operation	before	after
Filter::as_json	302 ns, 5 allocs	188 ns, 2 allocs
Filter::from_json	724 ns, 15 allocs	584 ns, 12 allocs

A filter with no generic tags is unchanged (55 -> 52 ns, 119 -> 116 ns, one allocation either way), since neither path allocated for keys in that case. The win is proportional to how many tag queries a filter carries.

Notes to the reviewers

Output is unchanged. Ten filter shapes, their round-trips, and fifteen well-formed and malformed inputs all produce byte-identical results, including which keys are ignored and which are rejected.

The key-shape rules had only partial coverage, and I got two of them wrong while writing the test, so they are now pinned explicitly:

• # followed by exactly one letter is a tag query.
• # followed by exactly one non-letter is rejected. This includes "##", which matches the #X shape and then fails the letter check.
• Anything else is ignored: "#ab", "#", "t", "".

Verified with the full just precommit set plus cargo build --workspace --all-targets and cargo clippy --all-targets -- -D warnings, including the no_std and wasm32 combinations.

That is the last of the near-term follow-ups from #1425. The remaining one is larger and I would rather discuss it before writing anything: Event::from_json spends 43 allocations on a twelve-tag event, 37 of them inside Tags, because Tag holds a Vec<String> and Tags a Vec<Tag>. That is a representation change to a public type, so I will open an issue with the numbers rather than a PR.

Checklist

• I followed the contribution guidelines
• I updated the relevant CHANGELOG.md (if applicable)
• I understand and can explain all code in this PR

[codecov]

Codecov Report

❌ Patch coverage is 93.75000% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
nostr/src/filter/mod.rs	93.75%	3 Missing ⚠️

📢 Thoughts on this report? Let us know!

[JSKitty]

Codecov: The three lines are the visitor's expecting, which serde only calls when constructing a type-mismatch error. JSON object keys are always strings, so it isn't reachable through Filter::from_json.

[yukibtc]

Thanks!