fix: drop the percentage free-space floor everywhere

[irzhywau]

Removes every free-space check that demanded a fixed share of the volume (10% or 15%) stay free.

Why

A share of capacity scales with the disk, not with the work. On a 460 GiB Mac the 10% rule demanded 46 GiB and blocked builds, tests, updates and model downloads; on large volumes it would demand terabytes. #123's review asked for this change to leave #123 and come as its own PR. #203 also drops the 10% floor, inside its cutover commit, and can take this version when it rebases.

What changes

• Each check now requires only that the bytes it is about to write fit in the available space. Existing exact-byte accounting, overflow and consistency checks, and fixed safety margins stay.
• Runtime: the update, model and content preparation, the IPFS provider's capacity check and the Runtime's recheck of it, Browser image preparation, hosted egress history, publication and release snapshots.
• Behaviour change: an update now proceeds as long as its own bytes fit, so it can fill a volume almost completely. If they don't fit it still refuses with "not enough free space for the update; previous release preserved".
• Tooling and CI: the release scripts and signer, image packaging, the media-tools build, update-hop comparison, Browser qualification, both workflows, and the local pre-push gate, which now always runs its Rust checks.
• Unchanged: the absolute minimums stay. Source-Home setup keeps 16 GiB and the media build keeps 4 GiB.
• Docs: AGENTS.md, TASKS.md, state.md, the content distribution contract and scripts/README.md drop the percentage. Dated audits are left as historical records.

Fixed 2 GiB reserve (5 October decision)

• One shared constant, FREE_SPACE_RESERVE_BYTES (2 GiB) in elastos-common, kept free on top of the bytes a user-facing write needs, so a disk is never filled to zero.
• require_update_space uses it, which covers update staging, support staging and the update controller's check (require_controller_update_space, formerly require_controller_disk_reserve).
• install.sh keeps the same value and checks the download folder and the install folder for the binary's size plus the reserve before writing. A Rust test fails if the two values ever differ.
• AGENTS.md keeps the housekeeping rule for development and staging volumes as a fixed, per-machine amount (ELASTOS_DEV_MIN_FREE_GB, default 15 GB), never a share of the volume.

Verification (local, macOS arm64, rebased on develop 0200582b)

• just test-elastos (same as CI, with provider binaries), run with Chat and People with Home: connect, approve contacts and recover (#86) #123's commits on top: 4076 passed, 0 failed. Clippy with -D warnings and cargo fmt --check are clean, and the new capsule lint passes for ipfs-provider.
• Script suites pass: install-bootstrap (56, including the new reserve test), prepare-release-platform (19), publish-platform-artifacts (18), ci-local-prepush (58), ci-release-policy (19), release-mac-capacity (14), product data.
• Known failures not caused by this PR: release-signer-test.py fails the same 2 tests on unchanged develop in this environment (also seen through release-platform-input-test.py, which reruns it), and update-hop-compare-test.py has one Python 3.14 tarfile fixture error in a test this PR does not touch.

[andersalm]

Decision (Anders, 5 October): drop the percentage floors as proposed, with one change: user-facing writes (update staging, installer, model and content downloads) keep one shared fixed reserve of 2 GiB on top of the bytes they write, so a disk is never filled to zero; require_update_space and the installer check use it. Keep AGENTS.md's 10% housekeeping rule for development and staging volumes. Please rebase after the update PRs (#243, #245, #246, #248) merge.

[andersalm]

Blocker for #108: the 2 GiB reserve from the 5 Oct decision is not in a7c23c3 yet (needed <= available). Please add one shared constant, used by require_update_space, require_controller_disk_reserve and install.sh.

[irzhywau]

@andersalm the requested changes have been pushed, just waiting for the CI to complete now

[andersalm]

@irzhywau as agreed, the operator took over this rebase. On top of your 0c67287:

• Merged current develop (da40d19): no text conflicts; one compile fix for fix(update): a release the publisher's signature rules out is not offered again; clear retry messages (#110) #257's test.
• One shared 2 GiB reserve, elastos_common::free_space (6049c8a). It covers update staging, the installer writer and backup copy, support staging, the update controller, model/content preparation, IPFS capacity, the Browser image and setup component installs (a3bc5a1). install.sh uses the same value; a parity test keeps them equal.
• AGENTS.md: the 10% housekeeping rule for development volumes is restored, per the 5 Oct decision.
• No percentage floor remains in product code.

Tests: boundary tests at exactly bytes + 2 GiB; 108 setup tests pass; the pre-push gate passed. GPT-6 Sol review: PASS after one gap (setup component writes) was fixed. It merges when CI is green. #123 now targets develop; I'll rebase it after this merges.