fix: bump deepagents floor to >=0.5.3 to unblock CI

[chrispatil]

What

CI on main is currently red (has been since commit 11fede7, 2026-08-05).
pyproject.toml pins deepagents>=0.4.1, but open_strix/app.py calls
CompositeBackend(..., artifacts_root=...) -- a kwarg that doesn't exist
until deepagents 0.5.x. A clean uv sync --dev resolves to 0.4.1 and every
test that constructs an OpenStrixApp fails at agent-build time with:

TypeError: CompositeBackend.__init__() got an unexpected keyword argument 'artifacts_root'

Confirmed independently via the GitHub Actions API: the Prerelease run for
main's current HEAD has conclusion: failure, with no follow-up fix commit
since. Reproduced locally on a pristine clone using CI's literal
uv sync --dev && pytest tests/ ... invocation: 86 failed, 252 passed.

The actual runtime this was caught in already has deepagents==0.5.3
installed (hand-patched outside of git at some point) -- the repo's own lock
file was never updated to match, so a fresh clone/CI run reproduces the
original break.

Fix

Bump the floor to deepagents>=0.5.3,<0.6 and regenerate the lockfile
(resolves to deepagents==0.5.9, matching what's already proven-good in
production). Also verified an unconstrained >=0.5.3 (resolves to current
latest 0.7.8) is fully green too, if a floating latest is preferred instead
of a pinned minor -- happy to switch.

Testing

Pristine clone, CI's exact pytest tests/ invocation (current 9-file ignore
list): 86 failed -> 0 failed, 252 passed -> 338 passed.

Why this matters for the other PRs in this batch

Every other branch needed this fix synced locally before its own tests could
even run -- OpenStrixApp() can't construct an agent without it. Recommend
merging this one first; the others should rebase onto it.

[chrispatil]

Closing as duplicate — #136 (filed 2026-08-05, still open) already fixes this exact break with a nearly identical diff (unbounded >=0.5.1 vs. this PR's >=0.5.3,<0.6; both verified green). Should have checked open PRs before filing. Redirecting the 3 dependent PRs in this batch (#138, #139, #140) to reference #136 instead.