fix: pass --no-input and bounded timeouts to pip

[ndonkoHenri]

Problem

When an index answers 401 — a corporate mirror, an authenticating proxy — pip prompts for credentials. The package command starts pip with a stdin pipe that nothing writes to and nothing closes, so the prompt can never be answered, and pip blocks forever. flet build shows this as a frozen Packaging Python app... with no error, no progress and no timeout.

The prompt is also invisible: pip writes User for <host>: to stdout with no trailing newline, and this command routes child stdout through verbose(), which flet build only enables at -vv. So the user gets a spinner and nothing else.

Reproduced against a local index returning 401 WWW-Authenticate: Basic, driving the same pip the tool bundles (26.1.2, from the pinned python-build-standalone distro), with stdin held open the way Process.start leaves it:

as on main:     HUNG — killed at 30s | stdout=76B stderr=0B
                last output: "User for 127.0.0.1:52836: "
with this PR:   rc=1 in 1.2s | stdout=50B stderr=135B
                last output: "ERROR: No matching distribution found for foo"

Change

--no-input makes pip fail with a readable error instead of prompting. --timeout 30 --retries 3 pins the network bounds on the command line so they hold regardless of a user's pip.conf or PIP_* environment (CLI args win over both). pip's own defaults are 15 s / 5 retries, so this is not adding bounds where there were none — it is trading two retries for twice the patience per attempt, which suits the slow and throttled links these reports come from, and makes the worst case independent of ambient config.

Refs flet-dev/flet#5989, flet-dev/flet#5013, flet-dev/flet#5507.

Scope

This closes one cause of the "packaging hangs" family, not all of them. Two things it deliberately does not cover:

• *A child that floods stderr still deadlocks.: runExec drains only stdout while the child runs, so a chatty or failing wheel build wedges on the unread stderr pipe — ~1 KB is enough on Windows. That is the other half of bug: flet build windows gets stuck on packaging python app; and extension template generation issues flet#5507 and will be thought of separately.
• Closing the child's stdin is the complementary primitive. Worth noting the two interact: with stdin closed but no --no-input, pip dies with a 9.9 KB EOFError traceback — which on Windows is itself large enough to wedge the undrained stderr pipe. --no-input is the better fix precisely because it never generates that output.