feat: booting indicator for async tool sources

[StanAngeloff]

Summary

When Flemma loads async tool sources (e.g. MCP servers, user-provided resolve functions), there's currently no way for users to know that tools are still loading. This PR adds visibility into the boot process through three surfaces:

• Lualine #{booting} variable — a new resolver that returns truthy while async tool sources are pending, allowing users to show a loading indicator in their statusline format string (e.g. #{?#{booting},⏳ ,}#{model})
• :Flemma status booting line — a ⏳ loading async tool sources… line appears in the Tools section while sources are still resolving
• FlemmaBootComplete autocmd — a User autocmd that fires once all async tool sources have resolved, enabling downstream integrations (the lualine component uses this to auto-refresh)

Architecture

No new modules. This extends three existing files plus the syntax file:

lua/flemma/tools/init.lua

fire_ready_callbacks() now emits vim.api.nvim_exec_autocmds("User", { pattern = "FlemmaBootComplete" }) after draining the callback queue. This is the single event that signals boot completion.

lua/lualine/components/flemma.lua

• New booting resolver in make_resolvers(): returns "1" while tools.is_ready() is false, "" when ready. This integrates with the existing conditional format system (#{?#{booting},true_branch,false_branch}).
• New init() override: creates a User FlemmaBootComplete autocmd that calls lualine.refresh(), so the statusline updates automatically when boot finishes (no polling needed).

lua/flemma/status.lua

• collect_tools() now includes booting = not tools_module.is_ready() in its return table.
• format() conditionally inserts ⏳ loading async tool sources… after the Tools header when data.tools.booting is true.
• The flemma.status.Data type annotation is updated to include booting: boolean on the tools field.

syntax/flemma_status.vim

• New FlemmaStatusBooting match for ^\s\+⏳ .*$, linked to WarningMsg highlight group.

Testing

All changes are TDD — tests were written first, verified to fail, then implementation was added:

• tests/flemma/tool_async_spec.lua — 1 new test: verifies FlemmaBootComplete autocmd fires when the last async source resolves
• tests/flemma/lualine_spec.lua — 3 new tests: #{booting} truthy while pending, falsy when ready, and lualine refresh on FlemmaBootComplete. The lualine mock was also improved to support proper init() inheritance (added init method and setmetatable-based extend)
• tests/flemma/status_spec.lua — 4 new tests: data.tools.booting true/false in collect(), and ⏳ presence/absence in format()

Test plan

• make qa passes (verified: qa: OK)
• New lualine variable works in a format string: #{?#{booting},⏳ loading…,}#{model}
• :Flemma status shows ⏳ loading async tool sources… while a slow async source is pending
• :Flemma status omits the booting line once all sources resolve
• Lualine auto-refreshes when boot completes (no manual refresh needed)

🤖 Generated with Claude Code

Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com