fix(cli): make string-valued ANSI mappings insert the mapped character

[salch-cred]

Fixes #92343

What & why

After #87511 mapped the Shift+letter CSI sequences in ANSI_SEQUENCES, the parser emits the right key — but Vt100Parser._call_handler always passes the raw matched byte sequence as KeyPress.data. The default self-insert binding inserts event.data, so Shift+letter still leaked [27;2;97~ into the prompt buffer on Ghostty.

Stock prompt_toolkit has zero plain-string mappings (all 180 string values in ANSI_SEQUENCES are Keys members — verified against 3.0.52), so the only plain-string values are Hermes' own Shift+letter table. That gives the fix a perfect blast radius:

install_vt100_str_key_data_fix() patches _call_handler so a plain-string mapped value becomes its own data payload. Keys members are str subclasses and are excluded explicitly — every Keys.* mapping keeps its raw-byte data. Wired into the existing startup install block in cli.py; idempotent.

How to test

scripts/run_tests.sh tests/hermes_cli/test_pt_input_extras_str_key_data.py -q

Tests drive a real Vt100Parser with Hermes' actual mappings installed: without the fix, feeding ESC[27;2;97~ yields KeyPress("A", "\x1b[27;2;97~") (the leak); with it, KeyPress("A", "A"). Also covered: Keys.* data payloads unchanged, install idempotency.

Manual repro: Ghostty + hermes → type Shift+M. Before: [27;2;109~ lands in the buffer. After: M.

Platforms tested

• Windows 11, Python 3.12: new suite 4/4.
• Parser semantics are host-independent; the reported terminal was macOS/Ghostty, but the data-payload defect is in the parser, not the terminal.

[alt-glitch]

This was generated by AI during triage.

Related to #87785 and #88097: all normalize raw VT100 KeyPress.data for character-valued aliases. #88097 covers a broader set of extended-key aliases; these are competing implementations.

[salch-cred]

Rebased on latest main — tests 4/4 passing. Thanks @isimoesg-amzn for the thorough independent verification (the A/B probe and the 384-test regression run are exactly the confidence this needed). The blast-radius argument holding on 3.0.52 is good to have on record.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #88097 — both patch Vt100Parser._call_handler to normalize raw CSI KeyPress.data; #88097 covers the broader key-alias set.

[salch-cred]

Saw the triage flag about #88097. My PR takes a different approach: instead of expanding the key-alias set, it patches the data payload at the parser level so ANY plain-string mapped value (not just the ones #88097 covers) inserts the mapped character. If #88097 is preferred, happy to consolidate — but the two fixes operate at different layers and could complement each other.