chore: split out non-A2A cleanups bundled in PR #272 (oracle_pool_cache + memory adapters)

[fede-kamel]

Tracking the scope-creep cleanup discussed in #272.

Two changes landed in the A2A v1 PR as "rebase fallout" but are unrelated to A2A v1.0 itself:

• src/locus/_oracle_pool_cache.py — tightened the generic typing on get_pool (TypeVar + cast). Improvement, but unrelated to A2A.
• src/locus/memory/backends/adapters.py — refactored two duplicated inspect.signature(...) blocks into a _accepts_parameter() helper. Improvement, but unrelated to A2A.

Both are good cleanups on their own; neither has anything to do with A2A v1.0. Bundling them into the v1 PR muddies the blame/bisect story if something later regresses in either subsystem.

What to do

Lift these two changes out of feat/a2a-v1-protocol-compat into their own small "chore" PR(s) targeting main. They can either:

• Ship as one PR titled something like chore: misc typing + signature-introspection cleanups
• Ship as two single-purpose PRs (one per file)

Both shapes are fine; the only ask is that they don't live under an A2A-v1 commit.

Why not block PR #272 on this

Mostly cosmetic — the changes are individually safe and Luigi already isolated them into their own commits on the branch (d4be37f, 8a9d013). So bisect within the branch is clean. The issue is just that once squash-merged, the history will record them under the A2A-v1 PR title.

If PR #272 gets squashed, the cleanest path is to land this issue's PRs first against main, then rebase #272 to pick up main (the rebase will drop those commits as already-applied).

If PR #272 gets merged with a merge commit (preserving the per-commit history), this issue can be closed without action — the commits speak for themselves.

Origin

Raised in this review on PR #272.

[luigisaetta]

@fede-kamel I think this one can be closed. IN the PR implementing support for A2A V1 the commit you mention here has been removed.

[fede-kamel]

Closing as a no-op — PR #272 merged as a true merge commit (c8b75bc), so d4be37f (fix(oracle): preserve pool cache return type) and 8a9d013 (fix(storage): tolerate mock backend signatures) are preserved on main with their original messages. Per this issue's own acceptance criterion: "If PR #272 gets merged with a merge commit (preserving the per-commit history), this issue can be closed without action — the commits speak for themselves."