feat(site): show total agent hours consumption on the licenses page - #27953

An enabled agent_runtime_hours feature with the limit omitted means the
license grants unlimited runtime hours (allocation claim -1). The card
previously rendered the invalid-limits error for that shape; it now
shows a full neutral bar with "Limit: Unlimited" and the tooltip
without threshold copy, mirroring SeatUsageBarCard's unlimited state.
The invalid-limits error is kept for negative limits, which can only
come from a decoding bug.

The allocation-reached tooltip now reports the floored actual usage
percentage (1,200 of 1,000 hours shows 120%) instead of always claiming
100%. The soft-limit percentage is floored rather than rounded so a
999/1,000-hour soft limit displays as 99% instead of crossing the
allocation boundary. Missing usage data no longer counts as reaching a
zero soft limit, and story tooltip assertions query the role=tooltip
node so they fail when the popover does not open.

A full green bar with no thresholds read as 100% usage. Unlimited
allocations now fill the track with purple diagonal stripes so the
full-width bar reads as an unmetered allocation instead of usage
approaching a limit.

The unlimited allocation bar keeps its diagonal stripes but switches
from purple back to the green highlight, and the right half of the
bar fades out into the track so the pattern reads as an open-ended
allocation rather than a full meter.

When a license carries a hard limit, the track scales to the
enforcement range: a solid red line marks the hard cap at the right
edge and a dotted yellow line marks the allocation, whose limit label
follows the marker. If the allocation sits close enough to the hard
cap for the labels to collide, the hard cap label moves above the
bar. Usage at or beyond the hard cap fills the track with red
diagonal stripes and the tooltip names the cap.

Threshold percentages now render with one floored decimal place
(999/1,000 hours shows 99.9% instead of 99%), still never crossing a
boundary the underlying hours have not reached.

…w total agent hours with one decimal

Derive the Total agent hours card's usage from the entitlement's new
actual_ms field, floored to tenths of an hour with integer math so the
displayed number and the reached states flip at the same instant as the
backend's whole-hour thresholds. Usage renders with exactly one decimal
(e.g. 435.8, 1,000.0); the limit and hard cap labels stay whole because
the claims are whole hours.

The fill is now colored by position instead of switching as a whole:
green until the soft limit, yellow from the soft limit to the
allocation, and red from the allocation to the hard cap. Each
threshold carries a marker line extending past the track: dotted
yellow for the soft limit, red for the allocation, and a double-width
primary-color line for the hard cap. Reaching the hard cap shows a
destructive pill above the bar stating that chat concurrency is
enforced.

New stories cover usage between the soft limit and the allocation and
between the allocation and the hard cap.

… agent hours labels inside the card

Replace the emdash fallback for missing usage data with N/A, and move
the limit label above the bar when the allocation marker sits within
the first 15% of a hard-cap scaled track, where the centered label
would spill outside the card and over the Used label.

…ect agent hours reached states

Turn the whole fill red when a metered allocation without a hard cap
is reached: the track ends at the allocation, so there is no position
past the limit where a red segment could render, and the previous
gating on a hard cap left the bar green or yellow in the reached
state. Also drop the concurrency-enforcement claim from the hard-cap
pill because no backend path enforces the hard limit yet.

Add MockAgentRuntimeHoursFeature to testHelpers/entities and spread it
in every TotalAgentHoursCard story instead of re-declaring the server
Feature shape per story, so entitlement defaults and the paired
actual/actual_ms fields stay defined in one place.

… reuse the runtime hours fixture in the view story

Spread MockAgentRuntimeHoursFeature in the TotalAgentHoursUsage view
story instead of rebuilding the entity inline, and drop the story
comment that restated the story name and its ordering assertion.

Below the md breakpoint the centered limit label collides with the
right-aligned hard cap label at interior marker positions, so it moves
above the bar instead, left-aligned in the row shared with the
hard-cap pill. Also drop the fixture comment that restated the
MockAgentRuntimeHoursFeature values.

… the bar red at a coincident hard cap

The backend accepts a hard cap equal to the allocation, which places
the shared threshold at the track's right edge where no red segment
can render past the limit. Key the full-red reached fill on the limit
boundary sitting at the track edge instead of on the absence of a hard
cap, so the coincident case reads as reached like the no-cap case.

Renders the usage period dates above the bar, matching ManagedAgentsConsumption, so admins can tell which license period the shown consumption covers. A missing or unparsable period omits the dates row; the MissingUsagePeriod story pins that fallback and the shared fixture gains a usage_period.

Also trims the narrated comments in the card and its stories down to the non-obvious invariants, per review.

… the hard cap badge and use reached wording

The shared Badge applies text-nowrap, so the full-sentence hard cap
badge overflowed narrow cards; this instance wraps instead. The badge
shows at usage equal to the hard cap (the boundary where the
concurrency cap re-engages), so it now reads "limit reached" rather
than "exceeded", which is false at equality.

Also replaces the en dash in the usage period label with a hyphen,
which broke CI at lint/emdash.