refactor: narrow file-level Pyright suppressions

[chengyongru]

Context

PR #5158 enables BasedPyright strict checking across nanobot/. The strict-clean baseline currently contains 31 file-level directives such as:

# pyright: reportMissingTypeStubs=false, reportPrivateUsage=false

A file-level directive also suppresses future diagnostics of that category elsewhere in the same module, so its blast radius grows as the file evolves.

Goal

Gradually replace file-level Pyright suppressions with the narrowest justified alternative while keeping runtime behavior unchanged.

Approach

• Inventory the 31 current file-level directives and group them by cause: missing third-party stubs, private API usage, heterogeneous tool overrides, generated/dynamic SDK surfaces, and genuinely dead-code diagnostics.
• Remove suppressions made obsolete by better annotations or public APIs.
• Prefer a typed adapter, Protocol, local stub, or narrow runtime normalizer when that improves the owning boundary.
• When suppression remains necessary, prefer a line-level # pyright: ignore[rule] with the smallest possible scope and a clear local reason.
• Split the work into reviewable subsystem-sized PRs; do not combine it with unrelated runtime refactors.

Acceptance criteria

• No net increase in file-level Pyright directives.
• Each retained file-level directive has a documented reason that cannot reasonably be scoped to a line.
• Converted suppressions remain BasedPyright strict-clean with no global diagnostic downgrade.
• Closest subsystem tests and the full configured CI matrix stay green.
• Progress is measurable against the initial baseline of 31 file-level directives.

Non-goals

• Removing justified line-level suppressions solely to reach zero ignores.
• Adding behavior changes to satisfy the checker.
• Maintaining third-party packages' complete type stubs inside nanobot.

Related: #5158