ci: reduce PR workflow latency by relocating coverage

[PLeVasseur]

Summary

• Move tarpaulin coverage out of PR Test jobs in bundled/unbundled workflows and run standard functional test paths in PR CI.
• Add a dedicated Nightly Coverage workflow (schedule, workflow_dispatch, and push to main) that runs tarpaulin once and uploads coverage artifacts.
• Remove duplicated no-feature lint/fmt checks from unbundled lint and add Rust dependency/build caching to bundled and unbundled compile-heavy jobs.
• Preserve existing required PR job names (Lint, Test, Build documentation) and keep obtain_and_build_vsomeip intact.

CI timing comparison

Comparison method

• Baseline: latest green runs from merged PR Upgrade transports to up-rust 0.9 compatibility (MSRV 1.88) #75 (21772997390 bundled, 21772997392 unbundled)
• Initial for this PR: first successful runs on PR ci: reduce PR workflow latency by relocating coverage #76 (21774091310 bundled, 21774091311 unbundled)
• Latest for this PR: current successful runs on PR ci: reduce PR workflow latency by relocating coverage #76 (21830035189 bundled, 21830035143 unbundled)

Initial-run comparison (PR #75 baseline -> PR #76 first successful run)

Metric	Baseline (PR #75)	Initial (PR #76)	Delta
Bundled workflow total	12m30s	13m16s	+6.1%
Bundled Lint	12m21s	12m50s	+3.9%
Bundled Test	9m18s	4m54s	-47.3%
Unbundled workflow total	9m28s	7m28s	-21.1%
Unbundled Lint	9m01s	6m45s	-25.1%
Unbundled Test	9m05s	4m48s	-47.2%

Latest-run improvement on this PR after iterative branch updates (initial -> latest)

Metric	Initial (PR #76)	Latest (PR #76)	Delta
Bundled workflow total	13m16s	4m21s	-67.2%
Bundled lint critical path	12m50s	2m53s*	-77.5%
Bundled Test	4m54s	4m10s	-15.0%
Unbundled workflow total	7m28s	4m27s	-40.4%
Unbundled Lint	6m45s	1m30s	-77.8%
Unbundled Test	4m48s	4m04s	-15.3%

• Latest bundled lint is split into shards and aggregated (Lint (feature build+clippy) 2m49s + Lint aggregator 4s; Lint (base build+clippy) runs in parallel).

Notes

• This is not a forced cold-cache benchmark; numbers reflect normal CI behavior with cache reuse.
• obtain_and_build_vsomeip remained a cache-hit in all compared runs (10s baseline, 15s initial, 11s latest), so the main gains are from CI workflow/test/lint path changes.

Run links

• Baseline bundled: https://github.com/eclipse-uprotocol/up-streamer-rust/actions/runs/21772997390
• Baseline unbundled: https://github.com/eclipse-uprotocol/up-streamer-rust/actions/runs/21772997392
• Initial bundled: https://github.com/eclipse-uprotocol/up-streamer-rust/actions/runs/21774091310
• Initial unbundled: https://github.com/eclipse-uprotocol/up-streamer-rust/actions/runs/21774091311
• Latest bundled: https://github.com/eclipse-uprotocol/up-streamer-rust/actions/runs/21830035189
• Latest unbundled: https://github.com/eclipse-uprotocol/up-streamer-rust/actions/runs/21830035143

[sophokles73]

@PLeVasseur Am I right in assuming that this should be rebased after #75 has been merged?

[PLeVasseur]

@PLeVasseur Am I right in assuming that this should be rebased after #75 has been merged?

Yep, exactly, will do that now

[PLeVasseur]

FYI -- rebased this now.

[sophokles73]

LGTM