Only clear steamapps before the last steamcmd retry

[MrRussLuther]

Fixes #220.

entry.sh runs rm -rf steamapps before every steamcmd retry, so a single failed first attempt costs a full re-download. On 5.0.0 I had steamcmd self-update, then app_update fail with state is 0x6 while no CS2 update was pending. Attempt 2 wiped the manifest and re-verified 73 GB, failed with 0x202, and attempt 3 wiped again and started from zero. Logs are in the issue.

The wipe is also a trap on its own. 0x202 is update paused, which for steamcmd means not enough disk space. With no manifest, steamcmd seems to budget free space for a full install instead of the delta, so on a volume with less free space than the whole game (mine had about 22 GB) the wipe itself produces the 0x202, and the next wipe throws the install away. sloppy-ninja's log in the issue has the same shape: verify to 99.85%, pause, wipe, download from scratch. What worked for them by hand was retrying with validation and the manifest kept.

So this keeps MAX_ATTEMPTS=3 and the wipe from c225ca3 (#142), but makes each retry escalate instead of wiping first:

• attempt 1: unchanged
• attempt 2: validate, manifest kept. Repairs corrupt or half-applied files with a delta download and never deletes anything. Verifying 73 GB took about 2 minutes here, against 12 for the download. The one side effect is that it reverts hand-edited game files like gameinfo.gi, which pre.sh already exists for.
• attempt 3: clear steamapps, then retry (what attempt 2 does today)

Each failure now also prints df -h for the install dir, so a 0x202 reads as a disk message in the log rather than a mystery code.

Tested the loop on its own with a stub steamcmd.sh that fails N times: attempt 2 passes validate with steamapps intact, attempt 3 removes it, STEAMAPPVALIDATE=1 still validates on attempt 1, and the non-zero exit code still propagates when all three fail. I've been running the plain-retry-then-wipe version on my server through a wrapper since the incident in the issue.

[MrRussLuther]

Heads up that I expanded this slightly after opening it. Rereading the logs in #220, the manifest wipe looks like the thing that turns a pause into a from-scratch download, so attempt 2 now validates with the manifest kept and only attempt 3 wipes. Details and the disk-space reasoning are in the PR description. Happy to drop back to the plain retry if you'd rather keep it minimal.