feat: add SBPP_BanPlayerBySteamId native for offline player bans

[Rushaway]

Ports upstream sbpp#1511 onto the srcdslab fork.

What it adds

SBPP_BanPlayerBySteamId(int iAdmin, const char[] steamId, const char[] name, int iTime, const char[] sReason) — a native for recording a ban in the SBPP database for a player who is no longer connected (TK managers, anti-cheat, and other automated systems that fire after the target disconnects). SBPP_BanPlayer requires a live client index; this native takes a STEAM_0:X:Y string and a display name instead.

Behaviour (unchanged from upstream):

• Throws a native error if steamId is not SteamID2 format.
• Runs a duplicate-check SELECT first and skips the INSERT if an active ban already exists.
• Falls back to a server-lookup subquery for sid when serverID == -1.
• Fires SBPP_OnBanPlayer with iTarget = -1 only after the INSERT succeeds.

sourcebanspp.inc gains the native declaration and MarkNativeAsOptional.

Fork adaptations

The fork's _bans INSERT statements carry an extra admin_name column populated by an IFNULL((SELECT user FROM %s_admins WHERE authid = ... OR authid REGEXP ...), '') subquery. The ported INSERT (both the serverID == -1 and registered-server branches) was adjusted to include that column and subquery, matching the fork's existing UTIL_InsertBan / Native_SBBanPlayer queries. No fork behaviour was reverted.

Notes

• No local SourcePawn compiler; relying on the plugin-build workflow.
• Upstream PR feat: add SBPP_BanPlayerBySteamId native for offline player bans sbpp/sourcebans-pp#1511 stays open; nothing upstream was modified.

🤖 Generated with Claude Code

[Rushaway]

Review: approve after fixes (pushed as 99b7e22)

The port itself is faithful and the fork adaptation is correct. I verified the parts most likely to go wrong in a port of this shape:

Correct as ported

• admin_name column adaptation is right in both branches. Column list is 11 wide (authid, name, created, ends, length, reason, aid, adminIp, admin_name, sid, country) and the VALUES list is 11 wide, in the same order — IFNULL((SELECT user ...)) sits between adminIp and sid, matching SelectAddbanCallback. Format specifiers vs. arguments: 16/16 for the serverID == -1 branch, 14/14 for the registered-server branch.
• Style matches the fork: FormatEx + %s with manual DB.Escape, not DB.Format.
• The duplicate-check SELECT is byte-identical to the existing one at sbpp_main.sp:723.
• adminAuth[8] slicing is correct for the '^STEAM_[0-9]:%s$' REGEXP, and the "STEAM_ID_SERVER" fallback slices to "_SERVER", which harmlessly matches nothing and falls through to the IFNULL defaults.
• DataPack lifetime is sound: pack is deleted on both the error and success paths of the select callback, fwdPack is read and deleted before the results == null check in the insert callback. No leak, no use-after-free.
• .inc: signature matches the implementation, MarkNativeAsOptional present, no BOM, no merge residue.

Fixed

1. Query buffer truncation (the fork-specific one). char query[1024] was inherited from upstream, but the extra admin_name sub-select adds a second IFNULL((SELECT user FROM %s_admins ...)) plus two more adminAuth expansions. Worst case is now ~1210 bytes (128-byte escaped authid, 64-byte escaped name, 256-byte escaped reason, four DatabasePrefix and four adminAuth expansions), which FormatEx truncates into invalid SQL — silently, since the failure only shows up as a logged query error. Grown to 2048.
2. reasonEscaped off-by-one. reason is 128 bytes, so the escape destination needs 128 * 2 + 1 = 257, not 256. SQL_EscapeString refuses to write anything when the destination is one byte short, so a 128-char reason of all quotes would have spliced an uninitialised stack buffer into the INSERT. Now sizeof(reason) * 2 + 1. (Note: the same off-by-one exists pre-PR at sbpp_main.sp:1246 and :1457 — not touched here.)
3. SteamID2 validation was prefix-only. strncmp(steamId, "STEAM_", 6) accepts "STEAM_junk", and since the INSERT slices authid[8] for the REGEXP, that produced a garbage ban row rather than an error. Replaced with UTIL_IsValidSteamID2(), a full STEAM_X:Y:Z check that short-circuits before reading past the terminator on every truncated prefix.
4. Negative iTime accepted. Stored a negative length and an ends timestamp in the past. Now rejected.
5. iAdmin unbounded. if (!admin || !IsClientInGame(admin)) faults on an out-of-range index, and g_sSteamIDs[admin] would read out of bounds. Now bounds-checked, and the Admin_Ban flag check from Native_SBBanPlayer is applied for consistency with the sibling native. (Flag this if you'd rather keep upstream's permissive behaviour — it's a deliberate deviation.)
6. Stale index leaking into the forward. A non-live iAdmin was still written to the DataPack and pushed to SBPP_OnBanPlayer. Now normalised to 0.
7. Style: use the callback's db handle instead of the global DB, make both callbacks public, drop the [SBPP] log prefix — matching SelectAddbanCallback / InsertAddbanCallback.
8. Docs: recorded the error conditions on the native and the iTarget = -1 contract on the SBPP_OnBanPlayer forward, so third-party subscribers know to range-check.

Not changed (noted for the record)

• The duplicate-check uses authid = '%s' rather than the universe-agnostic authid REGEXP '^STEAM_[0-9]:%s$' used by the connect-time check at :353. Worst case is a redundant row when the same account is stored under a different universe digit. sm_addban has the same behaviour, so this matches file convention.
• adminAuth is interpolated unescaped. It comes from GetClientAuthId / the literal "STEAM_ID_SERVER", never from user input, and every existing _bans INSERT in this file does the same.

Verification: compiled locally with spcomp 1.12.0.7253 — sbpp_main.sp, sbpp_sleuth.sp and sbpp_report.sp all build clean with zero warnings. CI on 99b7e22 is green (Compile SourceMod plugins: pass).