[Impeller][Vulkan] Always reset per-surface host buffers on frame submit (#194827)

Fixes https://github.com/flutter/flutter/issues/194406

`GPUSurfaceVulkanImpeller` constructs its own private `AiksContext` (and
`HostBuffer` ring) per surface, rather than sharing an `AiksContext`
across surfaces like `GPUSurfaceMetalImpeller` does via `IOSContext`.
Because `GPUSurfaceVulkanImpeller::AcquireFrame` previously passed
`surface_frame.submit_info().frame_boundary` as `reset_host_buffer` on
its `SurfaceContextVK` path, any surface submitted with `frame_boundary
= false` (such as Android HCPP overlay surfaces in
`AndroidExternalViewEmbedder2`) never called `HostBuffer::Reset()`, so
it never advanced its ring index (`frame_index_`) and kept appending new
blocks to `device_buffers_[frame_index_]` without bound while the other
ring entries stayed unused.

***------------------AI pr description------------------***:

### Why fix `GPUSurfaceVulkanImpeller` instead of
`AndroidExternalViewEmbedder2`?

A `HostBuffer` ring is owned by an `AiksContext` and must be advanced
once per `(AiksContext, frame)`:
- **Shared `AiksContext`** (`GPUSurfaceMetalImpeller` via `IOSContext`,
`EmbedderExternalViewEmbedder`): multiple surfaces in a single frame
encode into the same `AiksContext` and `HostBuffer`, so
`reset_host_buffer` must be gated on `submit_info().frame_boundary` to
reset only after the final surface in the frame group is submitted.
- **Per-surface `AiksContext`** (`GPUSurfaceGLImpeller`,
`GPUSurfaceVulkanImpeller`): each `Surface` owns a dedicated
`AiksContext` and `HostBuffer` that only receives one `RenderToTarget`
call per frame, so `reset_host_buffer` must always be `true` regardless
of whether other surfaces follow in the frame.
- **Consistency across per-surface paths**: Both `GPUSurfaceGLImpeller`
(`gpu_surface_gl_impeller.cc:123`) and `GPUSurfaceVulkanImpeller`'s own
non-swapchain codepath (`gpu_surface_vulkan_impeller.cc:203`) already
hardcode `/*reset_host_buffer=*/true` for this exact reason; only the
`SurfaceContextVK` swapchain branch at line 122 was still forwarding
`surface_frame.submit_info().frame_boundary`.

Keeping `overlay_frame->set_submit_info({.frame_boundary = false})` in
`AndroidExternalViewEmbedder2` preserves accurate frame-group boundary
semantics (the overlay surface is not the final surface submitted in the
frame), while fixing `GPUSurfaceVulkanImpeller` ensures any non-boundary
surface backed by `GPUSurfaceVulkanImpeller` properly rotates its
private `HostBuffer` ring.

### Safety

- **No buffer is reused while the GPU may still read it.** `HostBuffer`
only reuses a ring entry 4 resets after it was written. Each swapchain
`AcquireNextDrawable()` first waits on the fence of the frame submitted
2 frames earlier (`kMaxPendingPresents = 2`), so with one reset per
frame, that entry's GPU work has already completed. This is the same
guarantee the root surface already relies on, since it resets every
frame.
- **Performance (zero steady-state allocations):** `HostBuffer`
pre-allocates one 1 MB `DeviceBuffer` per ring slot
(`device_buffers_[0..3]`), and `Reset()` retains the blocks used by that
frame (`size() <= current_buffer_ + 1`) while rewinding `offset_ = 0`
and advancing `frame_index_`. With `Reset()` called once per frame, the
overlay reuses its pre-allocated 1 MB blocks in place with zero new
allocations, instead of calling `CreateBuffer()` for a new 1 MB block
every time cumulative draws across frames exceed 1 MB.

## Pre-launch Checklist

- [x] I read the [Contributor Guide] and followed the process outlined
there for submitting PRs.
- [x] I read the [Tree Hygiene] wiki page, which explains my
responsibilities.
- [x] I read and followed the [Flutter Style Guide], including [Features
we expect every widget to implement].
- [x] I signed the [CLA].
- [x] I listed at least one issue that this PR fixes in the description
above.
- [x] I updated/added relevant documentation (doc comments with `///`).
- [x] I added new tests to check the change I am making, or this PR is
[test-exempt].
- [x] I followed the [breaking change policy] and added [Data Driven
Fixes] where supported.
- [x] All existing and new tests are passing.

---------

Co-authored-by: Gray Mackall <mackall@google.com>